blob: 6aff93c98acaf6020eabd37f2e9b49a04135f187 [file] [log] [blame]
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001/* memcontrol.c - Memory Controller
2 *
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
5 *
Pavel Emelianov78fb7462008-02-07 00:13:51 -08006 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
8 *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08009 * Memory thresholds
10 * Copyright (C) 2009 Nokia Corporation
11 * Author: Kirill A. Shutemov
12 *
Balbir Singh8cdea7c2008-02-07 00:13:50 -080013 * This program is free software; you can redistribute it and/or modify
14 * it under the terms of the GNU General Public License as published by
15 * the Free Software Foundation; either version 2 of the License, or
16 * (at your option) any later version.
17 *
18 * This program is distributed in the hope that it will be useful,
19 * but WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21 * GNU General Public License for more details.
22 */
23
24#include <linux/res_counter.h>
25#include <linux/memcontrol.h>
26#include <linux/cgroup.h>
Pavel Emelianov78fb7462008-02-07 00:13:51 -080027#include <linux/mm.h>
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -080028#include <linux/hugetlb.h>
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -080029#include <linux/pagemap.h>
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080030#include <linux/smp.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080031#include <linux/page-flags.h>
Balbir Singh66e17072008-02-07 00:13:56 -080032#include <linux/backing-dev.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080033#include <linux/bit_spinlock.h>
34#include <linux/rcupdate.h>
Balbir Singhe2224322009-04-02 16:57:39 -070035#include <linux/limits.h>
Paul Gortmakerb9e15ba2011-05-26 16:00:52 -040036#include <linux/export.h>
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -080037#include <linux/mutex.h>
Balbir Singhf64c3f52009-09-23 15:56:37 -070038#include <linux/rbtree.h>
Balbir Singhb6ac57d2008-04-29 01:00:19 -070039#include <linux/slab.h>
Balbir Singh66e17072008-02-07 00:13:56 -080040#include <linux/swap.h>
Daisuke Nishimura02491442010-03-10 15:22:17 -080041#include <linux/swapops.h>
Balbir Singh66e17072008-02-07 00:13:56 -080042#include <linux/spinlock.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080043#include <linux/eventfd.h>
44#include <linux/sort.h>
Balbir Singh66e17072008-02-07 00:13:56 -080045#include <linux/fs.h>
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -080046#include <linux/seq_file.h>
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -070047#include <linux/vmalloc.h>
Christoph Lameterb69408e2008-10-18 20:26:14 -070048#include <linux/mm_inline.h>
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070049#include <linux/page_cgroup.h>
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -080050#include <linux/cpu.h>
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -070051#include <linux/oom.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080052#include "internal.h"
Balbir Singh8cdea7c2008-02-07 00:13:50 -080053
Balbir Singh8697d332008-02-07 00:13:59 -080054#include <asm/uaccess.h>
55
KOSAKI Motohirocc8e9702010-08-09 17:19:57 -070056#include <trace/events/vmscan.h>
57
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070058struct cgroup_subsys mem_cgroup_subsys __read_mostly;
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070059#define MEM_CGROUP_RECLAIM_RETRIES 5
Balbir Singh4b3bde42009-09-23 15:56:32 -070060struct mem_cgroup *root_mem_cgroup __read_mostly;
Balbir Singh8cdea7c2008-02-07 00:13:50 -080061
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080062#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
Li Zefan338c8432009-06-17 16:27:15 -070063/* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080064int do_swap_account __read_mostly;
Michal Hockoa42c3902010-11-24 12:57:08 -080065
66/* for remember boot option*/
67#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED
68static int really_do_swap_account __initdata = 1;
69#else
70static int really_do_swap_account __initdata = 0;
71#endif
72
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080073#else
74#define do_swap_account (0)
75#endif
76
77
Balbir Singh8cdea7c2008-02-07 00:13:50 -080078/*
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080079 * Statistics for memory cgroup.
80 */
81enum mem_cgroup_stat_index {
82 /*
83 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
84 */
85 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
Balbir Singhd69b0422009-06-17 16:26:34 -070086 MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */
KAMEZAWA Hiroyukid8046582009-12-15 16:47:09 -080087 MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */
Balbir Singh0c3e73e2009-09-23 15:56:42 -070088 MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -070089 MEM_CGROUP_STAT_DATA, /* end of data requires synchronization */
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -070090 MEM_CGROUP_ON_MOVE, /* someone is moving account between groups */
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080091 MEM_CGROUP_STAT_NSTATS,
92};
93
Johannes Weinere9f89742011-03-23 16:42:37 -070094enum mem_cgroup_events_index {
95 MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */
96 MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */
97 MEM_CGROUP_EVENTS_COUNT, /* # of pages paged in/out */
Ying Han456f9982011-05-26 16:25:38 -070098 MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */
99 MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */
Johannes Weinere9f89742011-03-23 16:42:37 -0700100 MEM_CGROUP_EVENTS_NSTATS,
101};
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700102/*
103 * Per memcg event counter is incremented at every pagein/pageout. With THP,
104 * it will be incremated by the number of pages. This counter is used for
105 * for trigger some periodic events. This is straightforward and better
106 * than using jiffies etc. to handle periodic memcg event.
107 */
108enum mem_cgroup_events_target {
109 MEM_CGROUP_TARGET_THRESH,
110 MEM_CGROUP_TARGET_SOFTLIMIT,
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700111 MEM_CGROUP_TARGET_NUMAINFO,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700112 MEM_CGROUP_NTARGETS,
113};
114#define THRESHOLDS_EVENTS_TARGET (128)
115#define SOFTLIMIT_EVENTS_TARGET (1024)
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700116#define NUMAINFO_EVENTS_TARGET (1024)
Johannes Weinere9f89742011-03-23 16:42:37 -0700117
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800118struct mem_cgroup_stat_cpu {
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700119 long count[MEM_CGROUP_STAT_NSTATS];
Johannes Weinere9f89742011-03-23 16:42:37 -0700120 unsigned long events[MEM_CGROUP_EVENTS_NSTATS];
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700121 unsigned long targets[MEM_CGROUP_NTARGETS];
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800122};
123
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800124/*
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800125 * per-zone information in memory controller.
126 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800127struct mem_cgroup_per_zone {
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -0800128 /*
129 * spin_lock to protect the per cgroup LRU
130 */
Christoph Lameterb69408e2008-10-18 20:26:14 -0700131 struct list_head lists[NR_LRU_LISTS];
132 unsigned long count[NR_LRU_LISTS];
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800133
134 struct zone_reclaim_stat reclaim_stat;
Balbir Singhf64c3f52009-09-23 15:56:37 -0700135 struct rb_node tree_node; /* RB tree node */
136 unsigned long long usage_in_excess;/* Set to the value by which */
137 /* the soft limit is exceeded*/
138 bool on_tree;
Balbir Singh4e416952009-09-23 15:56:39 -0700139 struct mem_cgroup *mem; /* Back pointer, we cannot */
140 /* use container_of */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800141};
142/* Macro for accessing counter */
143#define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
144
145struct mem_cgroup_per_node {
146 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
147};
148
149struct mem_cgroup_lru_info {
150 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
151};
152
153/*
Balbir Singhf64c3f52009-09-23 15:56:37 -0700154 * Cgroups above their limits are maintained in a RB-Tree, independent of
155 * their hierarchy representation
156 */
157
158struct mem_cgroup_tree_per_zone {
159 struct rb_root rb_root;
160 spinlock_t lock;
161};
162
163struct mem_cgroup_tree_per_node {
164 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
165};
166
167struct mem_cgroup_tree {
168 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
169};
170
171static struct mem_cgroup_tree soft_limit_tree __read_mostly;
172
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800173struct mem_cgroup_threshold {
174 struct eventfd_ctx *eventfd;
175 u64 threshold;
176};
177
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700178/* For threshold */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800179struct mem_cgroup_threshold_ary {
180 /* An array index points to threshold just below usage. */
Phil Carmody5407a562010-05-26 14:42:42 -0700181 int current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800182 /* Size of entries[] */
183 unsigned int size;
184 /* Array of thresholds */
185 struct mem_cgroup_threshold entries[0];
186};
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700187
188struct mem_cgroup_thresholds {
189 /* Primary thresholds array */
190 struct mem_cgroup_threshold_ary *primary;
191 /*
192 * Spare threshold array.
193 * This is needed to make mem_cgroup_unregister_event() "never fail".
194 * It must be able to store at least primary->size - 1 entries.
195 */
196 struct mem_cgroup_threshold_ary *spare;
197};
198
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700199/* for OOM */
200struct mem_cgroup_eventfd_list {
201 struct list_head list;
202 struct eventfd_ctx *eventfd;
203};
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800204
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700205static void mem_cgroup_threshold(struct mem_cgroup *memcg);
206static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800207
Balbir Singhf64c3f52009-09-23 15:56:37 -0700208/*
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800209 * The memory controller data structure. The memory controller controls both
210 * page cache and RSS per cgroup. We would eventually like to provide
211 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
212 * to help the administrator determine what knobs to tune.
213 *
214 * TODO: Add a water mark for the memory controller. Reclaim will begin when
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800215 * we hit the water mark. May be even add a low water mark, such that
216 * no reclaim occurs from a cgroup at it's low water mark, this is
217 * a feature that will be implemented much later in the future.
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800218 */
219struct mem_cgroup {
220 struct cgroup_subsys_state css;
221 /*
222 * the counter to account for memory usage
223 */
224 struct res_counter res;
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800225 /*
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800226 * the counter to account for mem+swap usage.
227 */
228 struct res_counter memsw;
229 /*
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800230 * Per cgroup active and inactive list, similar to the
231 * per zone LRU lists.
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800232 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800233 struct mem_cgroup_lru_info info;
Balbir Singh6d61ef42009-01-07 18:08:06 -0800234 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -0200235 * While reclaiming in a hierarchy, we cache the last child we
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -0700236 * reclaimed from.
Balbir Singh6d61ef42009-01-07 18:08:06 -0800237 */
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -0700238 int last_scanned_child;
Ying Han889976d2011-05-26 16:25:33 -0700239 int last_scanned_node;
240#if MAX_NUMNODES > 1
241 nodemask_t scan_nodes;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700242 atomic_t numainfo_events;
243 atomic_t numainfo_updating;
Ying Han889976d2011-05-26 16:25:33 -0700244#endif
Balbir Singh18f59ea2009-01-07 18:08:07 -0800245 /*
246 * Should the accounting and control be hierarchical, per subtree?
247 */
248 bool use_hierarchy;
Michal Hocko79dfdac2011-07-26 16:08:23 -0700249
250 bool oom_lock;
251 atomic_t under_oom;
252
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800253 atomic_t refcnt;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -0800254
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -0700255 int swappiness;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -0700256 /* OOM-Killer disable */
257 int oom_kill_disable;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -0800258
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -0700259 /* set when res.limit == memsw.limit */
260 bool memsw_is_minimum;
261
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800262 /* protect arrays of thresholds */
263 struct mutex thresholds_lock;
264
265 /* thresholds for memory usage. RCU-protected */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700266 struct mem_cgroup_thresholds thresholds;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700267
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800268 /* thresholds for mem+swap usage. RCU-protected */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700269 struct mem_cgroup_thresholds memsw_thresholds;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700270
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700271 /* For oom notifier event fd */
272 struct list_head oom_notify;
Johannes Weiner185efc02011-09-14 16:21:58 -0700273
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800274 /*
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800275 * Should we move charges of a task when a task is moved into this
276 * mem_cgroup ? And what type of charges should we move ?
277 */
278 unsigned long move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800279 /*
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800280 * percpu counter.
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800281 */
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800282 struct mem_cgroup_stat_cpu *stat;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700283 /*
284 * used when a cpu is offlined or other synchronizations
285 * See mem_cgroup_read_stat().
286 */
287 struct mem_cgroup_stat_cpu nocpu_base;
288 spinlock_t pcp_counter_lock;
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800289};
290
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800291/* Stuffs for move charges at task migration. */
292/*
293 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
294 * left-shifted bitmap of these types.
295 */
296enum move_type {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800297 MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700298 MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800299 NR_MOVE_TYPE,
300};
301
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800302/* "mc" and its members are protected by cgroup_mutex */
303static struct move_charge_struct {
Daisuke Nishimurab1dd6932010-11-24 12:57:06 -0800304 spinlock_t lock; /* for from, to */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800305 struct mem_cgroup *from;
306 struct mem_cgroup *to;
307 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800308 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800309 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800310 struct task_struct *moving_task; /* a task moving charges */
311 wait_queue_head_t waitq; /* a waitq for other context */
312} mc = {
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -0700313 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800314 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
315};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800316
Daisuke Nishimura90254a62010-05-26 14:42:38 -0700317static bool move_anon(void)
318{
319 return test_bit(MOVE_CHARGE_TYPE_ANON,
320 &mc.to->move_charge_at_immigrate);
321}
322
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700323static bool move_file(void)
324{
325 return test_bit(MOVE_CHARGE_TYPE_FILE,
326 &mc.to->move_charge_at_immigrate);
327}
328
Balbir Singh4e416952009-09-23 15:56:39 -0700329/*
330 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
331 * limit reclaim to prevent infinite loops, if they ever occur.
332 */
333#define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
334#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
335
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800336enum charge_type {
337 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
338 MEM_CGROUP_CHARGE_TYPE_MAPPED,
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700339 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700340 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800341 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -0700342 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700343 NR_CHARGE_TYPE,
344};
345
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800346/* for encoding cft->private value on file */
347#define _MEM (0)
348#define _MEMSWAP (1)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700349#define _OOM_TYPE (2)
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800350#define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
351#define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
352#define MEMFILE_ATTR(val) ((val) & 0xffff)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700353/* Used for OOM nofiier */
354#define OOM_CONTROL (0)
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800355
Balbir Singh75822b42009-09-23 15:56:38 -0700356/*
357 * Reclaim flags for mem_cgroup_hierarchical_reclaim
358 */
359#define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
360#define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
361#define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
362#define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
Balbir Singh4e416952009-09-23 15:56:39 -0700363#define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2
364#define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT)
Balbir Singh75822b42009-09-23 15:56:38 -0700365
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700366static void mem_cgroup_get(struct mem_cgroup *memcg);
367static void mem_cgroup_put(struct mem_cgroup *memcg);
368static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg);
369static void drain_all_stock_async(struct mem_cgroup *memcg);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800370
Balbir Singhf64c3f52009-09-23 15:56:37 -0700371static struct mem_cgroup_per_zone *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700372mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700373{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700374 return &memcg->info.nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700375}
376
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700377struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg)
Wu Fengguangd3242362009-12-16 12:19:59 +0100378{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700379 return &memcg->css;
Wu Fengguangd3242362009-12-16 12:19:59 +0100380}
381
Balbir Singhf64c3f52009-09-23 15:56:37 -0700382static struct mem_cgroup_per_zone *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700383page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700384{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700385 int nid = page_to_nid(page);
386 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700387
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700388 return mem_cgroup_zoneinfo(memcg, nid, zid);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700389}
390
391static struct mem_cgroup_tree_per_zone *
392soft_limit_tree_node_zone(int nid, int zid)
393{
394 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
395}
396
397static struct mem_cgroup_tree_per_zone *
398soft_limit_tree_from_page(struct page *page)
399{
400 int nid = page_to_nid(page);
401 int zid = page_zonenum(page);
402
403 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
404}
405
406static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700407__mem_cgroup_insert_exceeded(struct mem_cgroup *memcg,
Balbir Singhf64c3f52009-09-23 15:56:37 -0700408 struct mem_cgroup_per_zone *mz,
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700409 struct mem_cgroup_tree_per_zone *mctz,
410 unsigned long long new_usage_in_excess)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700411{
412 struct rb_node **p = &mctz->rb_root.rb_node;
413 struct rb_node *parent = NULL;
414 struct mem_cgroup_per_zone *mz_node;
415
416 if (mz->on_tree)
417 return;
418
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700419 mz->usage_in_excess = new_usage_in_excess;
420 if (!mz->usage_in_excess)
421 return;
Balbir Singhf64c3f52009-09-23 15:56:37 -0700422 while (*p) {
423 parent = *p;
424 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
425 tree_node);
426 if (mz->usage_in_excess < mz_node->usage_in_excess)
427 p = &(*p)->rb_left;
428 /*
429 * We can't avoid mem cgroups that are over their soft
430 * limit by the same amount
431 */
432 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
433 p = &(*p)->rb_right;
434 }
435 rb_link_node(&mz->tree_node, parent, p);
436 rb_insert_color(&mz->tree_node, &mctz->rb_root);
437 mz->on_tree = true;
Balbir Singh4e416952009-09-23 15:56:39 -0700438}
439
440static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700441__mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
Balbir Singh4e416952009-09-23 15:56:39 -0700442 struct mem_cgroup_per_zone *mz,
443 struct mem_cgroup_tree_per_zone *mctz)
444{
445 if (!mz->on_tree)
446 return;
447 rb_erase(&mz->tree_node, &mctz->rb_root);
448 mz->on_tree = false;
449}
450
451static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700452mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
Balbir Singhf64c3f52009-09-23 15:56:37 -0700453 struct mem_cgroup_per_zone *mz,
454 struct mem_cgroup_tree_per_zone *mctz)
455{
456 spin_lock(&mctz->lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700457 __mem_cgroup_remove_exceeded(memcg, mz, mctz);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700458 spin_unlock(&mctz->lock);
459}
460
Balbir Singhf64c3f52009-09-23 15:56:37 -0700461
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700462static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700463{
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700464 unsigned long long excess;
Balbir Singhf64c3f52009-09-23 15:56:37 -0700465 struct mem_cgroup_per_zone *mz;
466 struct mem_cgroup_tree_per_zone *mctz;
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700467 int nid = page_to_nid(page);
468 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700469 mctz = soft_limit_tree_from_page(page);
470
471 /*
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700472 * Necessary to update all ancestors when hierarchy is used.
473 * because their event counter is not touched.
Balbir Singhf64c3f52009-09-23 15:56:37 -0700474 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700475 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
476 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
477 excess = res_counter_soft_limit_excess(&memcg->res);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700478 /*
479 * We have to update the tree if mz is on RB-tree or
480 * mem is over its softlimit.
481 */
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700482 if (excess || mz->on_tree) {
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700483 spin_lock(&mctz->lock);
484 /* if on-tree, remove it */
485 if (mz->on_tree)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700486 __mem_cgroup_remove_exceeded(memcg, mz, mctz);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700487 /*
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700488 * Insert again. mz->usage_in_excess will be updated.
489 * If excess is 0, no tree ops.
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700490 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700491 __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700492 spin_unlock(&mctz->lock);
493 }
Balbir Singhf64c3f52009-09-23 15:56:37 -0700494 }
495}
496
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700497static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700498{
499 int node, zone;
500 struct mem_cgroup_per_zone *mz;
501 struct mem_cgroup_tree_per_zone *mctz;
502
503 for_each_node_state(node, N_POSSIBLE) {
504 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700505 mz = mem_cgroup_zoneinfo(memcg, node, zone);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700506 mctz = soft_limit_tree_node_zone(node, zone);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700507 mem_cgroup_remove_exceeded(memcg, mz, mctz);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700508 }
509 }
510}
511
Balbir Singh4e416952009-09-23 15:56:39 -0700512static struct mem_cgroup_per_zone *
513__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
514{
515 struct rb_node *rightmost = NULL;
KAMEZAWA Hiroyuki26251ea2009-10-01 15:44:08 -0700516 struct mem_cgroup_per_zone *mz;
Balbir Singh4e416952009-09-23 15:56:39 -0700517
518retry:
KAMEZAWA Hiroyuki26251ea2009-10-01 15:44:08 -0700519 mz = NULL;
Balbir Singh4e416952009-09-23 15:56:39 -0700520 rightmost = rb_last(&mctz->rb_root);
521 if (!rightmost)
522 goto done; /* Nothing to reclaim from */
523
524 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
525 /*
526 * Remove the node now but someone else can add it back,
527 * we will to add it back at the end of reclaim to its correct
528 * position in the tree.
529 */
530 __mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
531 if (!res_counter_soft_limit_excess(&mz->mem->res) ||
532 !css_tryget(&mz->mem->css))
533 goto retry;
534done:
535 return mz;
536}
537
538static struct mem_cgroup_per_zone *
539mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
540{
541 struct mem_cgroup_per_zone *mz;
542
543 spin_lock(&mctz->lock);
544 mz = __mem_cgroup_largest_soft_limit_node(mctz);
545 spin_unlock(&mctz->lock);
546 return mz;
547}
548
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700549/*
550 * Implementation Note: reading percpu statistics for memcg.
551 *
552 * Both of vmstat[] and percpu_counter has threshold and do periodic
553 * synchronization to implement "quick" read. There are trade-off between
554 * reading cost and precision of value. Then, we may have a chance to implement
555 * a periodic synchronizion of counter in memcg's counter.
556 *
557 * But this _read() function is used for user interface now. The user accounts
558 * memory usage by memory cgroup and he _always_ requires exact value because
559 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
560 * have to visit all online cpus and make sum. So, for now, unnecessary
561 * synchronization is not implemented. (just implemented for cpu hotplug)
562 *
563 * If there are kernel internal actions which can make use of some not-exact
564 * value, and reading all cpu value can be performance bottleneck in some
565 * common workload, threashold and synchonization as vmstat[] should be
566 * implemented.
567 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700568static long mem_cgroup_read_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700569 enum mem_cgroup_stat_index idx)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800570{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700571 long val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800572 int cpu;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800573
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700574 get_online_cpus();
575 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700576 val += per_cpu(memcg->stat->count[idx], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700577#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700578 spin_lock(&memcg->pcp_counter_lock);
579 val += memcg->nocpu_base.count[idx];
580 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700581#endif
582 put_online_cpus();
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800583 return val;
584}
585
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700586static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700587 bool charge)
588{
589 int val = (charge) ? 1 : -1;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700590 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAPOUT], val);
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700591}
592
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700593void mem_cgroup_pgfault(struct mem_cgroup *memcg, int val)
Ying Han456f9982011-05-26 16:25:38 -0700594{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700595 this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT], val);
Ying Han456f9982011-05-26 16:25:38 -0700596}
597
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700598void mem_cgroup_pgmajfault(struct mem_cgroup *memcg, int val)
Ying Han456f9982011-05-26 16:25:38 -0700599{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700600 this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT], val);
Ying Han456f9982011-05-26 16:25:38 -0700601}
602
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700603static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
Johannes Weinere9f89742011-03-23 16:42:37 -0700604 enum mem_cgroup_events_index idx)
605{
606 unsigned long val = 0;
607 int cpu;
608
609 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700610 val += per_cpu(memcg->stat->events[idx], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700611#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700612 spin_lock(&memcg->pcp_counter_lock);
613 val += memcg->nocpu_base.events[idx];
614 spin_unlock(&memcg->pcp_counter_lock);
Johannes Weinere9f89742011-03-23 16:42:37 -0700615#endif
616 return val;
617}
618
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700619static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800620 bool file, int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800621{
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800622 preempt_disable();
623
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800624 if (file)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700625 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
626 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800627 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700628 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
629 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700630
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800631 /* pagein of a big page is an event. So, ignore page size */
632 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700633 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800634 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700635 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800636 nr_pages = -nr_pages; /* for event */
637 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800638
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700639 __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_COUNT], nr_pages);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800640
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800641 preempt_enable();
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800642}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800643
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700644unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700645mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700646 unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700647{
648 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700649 enum lru_list l;
650 unsigned long ret = 0;
651
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700652 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700653
654 for_each_lru(l) {
655 if (BIT(l) & lru_mask)
656 ret += MEM_CGROUP_ZSTAT(mz, l);
657 }
658 return ret;
659}
660
661static unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700662mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700663 int nid, unsigned int lru_mask)
664{
Ying Han889976d2011-05-26 16:25:33 -0700665 u64 total = 0;
666 int zid;
667
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700668 for (zid = 0; zid < MAX_NR_ZONES; zid++)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700669 total += mem_cgroup_zone_nr_lru_pages(memcg,
670 nid, zid, lru_mask);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700671
Ying Han889976d2011-05-26 16:25:33 -0700672 return total;
673}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700674
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700675static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700676 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800677{
Ying Han889976d2011-05-26 16:25:33 -0700678 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800679 u64 total = 0;
680
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700681 for_each_node_state(nid, N_HIGH_MEMORY)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700682 total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800683 return total;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800684}
685
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700686static bool __memcg_event_check(struct mem_cgroup *memcg, int target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800687{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700688 unsigned long val, next;
689
Steven Rostedt47994012011-11-02 13:38:33 -0700690 val = __this_cpu_read(memcg->stat->events[MEM_CGROUP_EVENTS_COUNT]);
691 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700692 /* from time_after() in jiffies.h */
693 return ((long)next - (long)val < 0);
694}
695
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700696static void __mem_cgroup_target_update(struct mem_cgroup *memcg, int target)
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700697{
698 unsigned long val, next;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800699
Steven Rostedt47994012011-11-02 13:38:33 -0700700 val = __this_cpu_read(memcg->stat->events[MEM_CGROUP_EVENTS_COUNT]);
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800701
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700702 switch (target) {
703 case MEM_CGROUP_TARGET_THRESH:
704 next = val + THRESHOLDS_EVENTS_TARGET;
705 break;
706 case MEM_CGROUP_TARGET_SOFTLIMIT:
707 next = val + SOFTLIMIT_EVENTS_TARGET;
708 break;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700709 case MEM_CGROUP_TARGET_NUMAINFO:
710 next = val + NUMAINFO_EVENTS_TARGET;
711 break;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700712 default:
713 return;
714 }
715
Steven Rostedt47994012011-11-02 13:38:33 -0700716 __this_cpu_write(memcg->stat->targets[target], next);
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800717}
718
719/*
720 * Check events in order.
721 *
722 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700723static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800724{
Steven Rostedt47994012011-11-02 13:38:33 -0700725 preempt_disable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800726 /* threshold event is triggered in finer grain than soft limit */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700727 if (unlikely(__memcg_event_check(memcg, MEM_CGROUP_TARGET_THRESH))) {
728 mem_cgroup_threshold(memcg);
729 __mem_cgroup_target_update(memcg, MEM_CGROUP_TARGET_THRESH);
730 if (unlikely(__memcg_event_check(memcg,
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700731 MEM_CGROUP_TARGET_SOFTLIMIT))) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700732 mem_cgroup_update_tree(memcg, page);
733 __mem_cgroup_target_update(memcg,
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700734 MEM_CGROUP_TARGET_SOFTLIMIT);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700735 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700736#if MAX_NUMNODES > 1
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700737 if (unlikely(__memcg_event_check(memcg,
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700738 MEM_CGROUP_TARGET_NUMAINFO))) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700739 atomic_inc(&memcg->numainfo_events);
740 __mem_cgroup_target_update(memcg,
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700741 MEM_CGROUP_TARGET_NUMAINFO);
742 }
743#endif
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800744 }
Steven Rostedt47994012011-11-02 13:38:33 -0700745 preempt_enable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800746}
747
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800748static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800749{
750 return container_of(cgroup_subsys_state(cont,
751 mem_cgroup_subsys_id), struct mem_cgroup,
752 css);
753}
754
Balbir Singhcf475ad2008-04-29 01:00:16 -0700755struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800756{
Balbir Singh31a78f22008-09-28 23:09:31 +0100757 /*
758 * mm_update_next_owner() may clear mm->owner to NULL
759 * if it races with swapoff, page migration, etc.
760 * So this can be called with p == NULL.
761 */
762 if (unlikely(!p))
763 return NULL;
764
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800765 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
766 struct mem_cgroup, css);
767}
768
KOSAKI Motohiroa4336582011-06-15 15:08:13 -0700769struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800770{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700771 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700772
773 if (!mm)
774 return NULL;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800775 /*
776 * Because we have no locks, mm->owner's may be being moved to other
777 * cgroup. We use css_tryget() here even if this looks
778 * pessimistic (rather than adding locks here).
779 */
780 rcu_read_lock();
781 do {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700782 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
783 if (unlikely(!memcg))
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800784 break;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700785 } while (!css_tryget(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800786 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700787 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800788}
789
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700790/* The caller has to guarantee "mem" exists before calling this */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700791static struct mem_cgroup *mem_cgroup_start_loop(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700792{
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700793 struct cgroup_subsys_state *css;
794 int found;
795
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700796 if (!memcg) /* ROOT cgroup has the smallest ID */
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700797 return root_mem_cgroup; /*css_put/get against root is ignored*/
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700798 if (!memcg->use_hierarchy) {
799 if (css_tryget(&memcg->css))
800 return memcg;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700801 return NULL;
802 }
803 rcu_read_lock();
804 /*
805 * searching a memory cgroup which has the smallest ID under given
806 * ROOT cgroup. (ID >= 1)
807 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700808 css = css_get_next(&mem_cgroup_subsys, 1, &memcg->css, &found);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700809 if (css && css_tryget(css))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700810 memcg = container_of(css, struct mem_cgroup, css);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700811 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700812 memcg = NULL;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700813 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700814 return memcg;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700815}
816
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700817static struct mem_cgroup *mem_cgroup_get_next(struct mem_cgroup *iter,
818 struct mem_cgroup *root,
819 bool cond)
820{
821 int nextid = css_id(&iter->css) + 1;
822 int found;
823 int hierarchy_used;
824 struct cgroup_subsys_state *css;
825
826 hierarchy_used = iter->use_hierarchy;
827
828 css_put(&iter->css);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700829 /* If no ROOT, walk all, ignore hierarchy */
830 if (!cond || (root && !hierarchy_used))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700831 return NULL;
832
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700833 if (!root)
834 root = root_mem_cgroup;
835
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700836 do {
837 iter = NULL;
838 rcu_read_lock();
839
840 css = css_get_next(&mem_cgroup_subsys, nextid,
841 &root->css, &found);
842 if (css && css_tryget(css))
843 iter = container_of(css, struct mem_cgroup, css);
844 rcu_read_unlock();
845 /* If css is NULL, no more cgroups will be found */
846 nextid = found + 1;
847 } while (css && !iter);
848
849 return iter;
850}
851/*
852 * for_eacn_mem_cgroup_tree() for visiting all cgroup under tree. Please
853 * be careful that "break" loop is not allowed. We have reference count.
854 * Instead of that modify "cond" to be false and "continue" to exit the loop.
855 */
856#define for_each_mem_cgroup_tree_cond(iter, root, cond) \
857 for (iter = mem_cgroup_start_loop(root);\
858 iter != NULL;\
859 iter = mem_cgroup_get_next(iter, root, cond))
860
861#define for_each_mem_cgroup_tree(iter, root) \
862 for_each_mem_cgroup_tree_cond(iter, root, true)
863
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700864#define for_each_mem_cgroup_all(iter) \
865 for_each_mem_cgroup_tree_cond(iter, NULL, true)
866
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700867
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700868static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
Balbir Singh4b3bde42009-09-23 15:56:32 -0700869{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700870 return (memcg == root_mem_cgroup);
Balbir Singh4b3bde42009-09-23 15:56:32 -0700871}
872
Ying Han456f9982011-05-26 16:25:38 -0700873void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
874{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700875 struct mem_cgroup *memcg;
Ying Han456f9982011-05-26 16:25:38 -0700876
877 if (!mm)
878 return;
879
880 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700881 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
882 if (unlikely(!memcg))
Ying Han456f9982011-05-26 16:25:38 -0700883 goto out;
884
885 switch (idx) {
886 case PGMAJFAULT:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700887 mem_cgroup_pgmajfault(memcg, 1);
Ying Han456f9982011-05-26 16:25:38 -0700888 break;
889 case PGFAULT:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700890 mem_cgroup_pgfault(memcg, 1);
Ying Han456f9982011-05-26 16:25:38 -0700891 break;
892 default:
893 BUG();
894 }
895out:
896 rcu_read_unlock();
897}
898EXPORT_SYMBOL(mem_cgroup_count_vm_event);
899
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800900/*
901 * Following LRU functions are allowed to be used without PCG_LOCK.
902 * Operations are called by routine of global LRU independently from memcg.
903 * What we have to take care of here is validness of pc->mem_cgroup.
904 *
905 * Changes to pc->mem_cgroup happens when
906 * 1. charge
907 * 2. moving account
908 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
909 * It is added to LRU before charge.
910 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
911 * When moving account, the page is not on LRU. It's isolated.
912 */
913
914void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800915{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800916 struct page_cgroup *pc;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800917 struct mem_cgroup_per_zone *mz;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700918
Hirokazu Takahashif8d66542009-01-07 18:08:02 -0800919 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800920 return;
921 pc = lookup_page_cgroup(page);
922 /* can happen while we handle swapcache. */
Balbir Singh4b3bde42009-09-23 15:56:32 -0700923 if (!TestClearPageCgroupAcctLRU(pc))
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800924 return;
Balbir Singh4b3bde42009-09-23 15:56:32 -0700925 VM_BUG_ON(!pc->mem_cgroup);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -0800926 /*
927 * We don't check PCG_USED bit. It's cleared when the "page" is finally
928 * removed from global LRU.
929 */
Johannes Weiner97a6c372011-03-23 16:42:27 -0700930 mz = page_cgroup_zoneinfo(pc->mem_cgroup, page);
KAMEZAWA Hiroyukiece35ca2011-01-20 14:44:24 -0800931 /* huge page split is done under lru_lock. so, we have no races. */
932 MEM_CGROUP_ZSTAT(mz, lru) -= 1 << compound_order(page);
Balbir Singh4b3bde42009-09-23 15:56:32 -0700933 if (mem_cgroup_is_root(pc->mem_cgroup))
934 return;
935 VM_BUG_ON(list_empty(&pc->lru));
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800936 list_del_init(&pc->lru);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800937}
938
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800939void mem_cgroup_del_lru(struct page *page)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800940{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800941 mem_cgroup_del_lru_list(page, page_lru(page));
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800942}
943
Minchan Kim3f58a822011-03-22 16:32:53 -0700944/*
945 * Writeback is about to end against a page which has been marked for immediate
946 * reclaim. If it still appears to be reclaimable, move it to the tail of the
947 * inactive list.
948 */
949void mem_cgroup_rotate_reclaimable_page(struct page *page)
950{
951 struct mem_cgroup_per_zone *mz;
952 struct page_cgroup *pc;
953 enum lru_list lru = page_lru(page);
954
955 if (mem_cgroup_disabled())
956 return;
957
958 pc = lookup_page_cgroup(page);
959 /* unused or root page is not rotated. */
960 if (!PageCgroupUsed(pc))
961 return;
962 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
963 smp_rmb();
964 if (mem_cgroup_is_root(pc->mem_cgroup))
965 return;
Johannes Weiner97a6c372011-03-23 16:42:27 -0700966 mz = page_cgroup_zoneinfo(pc->mem_cgroup, page);
Minchan Kim3f58a822011-03-22 16:32:53 -0700967 list_move_tail(&pc->lru, &mz->lists[lru]);
968}
969
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800970void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
Balbir Singh66e17072008-02-07 00:13:56 -0800971{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800972 struct mem_cgroup_per_zone *mz;
973 struct page_cgroup *pc;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800974
Hirokazu Takahashif8d66542009-01-07 18:08:02 -0800975 if (mem_cgroup_disabled())
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700976 return;
Christoph Lameterb69408e2008-10-18 20:26:14 -0700977
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800978 pc = lookup_page_cgroup(page);
Balbir Singh4b3bde42009-09-23 15:56:32 -0700979 /* unused or root page is not rotated. */
Johannes Weiner713735b2011-01-20 14:44:31 -0800980 if (!PageCgroupUsed(pc))
981 return;
982 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
983 smp_rmb();
984 if (mem_cgroup_is_root(pc->mem_cgroup))
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800985 return;
Johannes Weiner97a6c372011-03-23 16:42:27 -0700986 mz = page_cgroup_zoneinfo(pc->mem_cgroup, page);
Christoph Lameterb69408e2008-10-18 20:26:14 -0700987 list_move(&pc->lru, &mz->lists[lru]);
Balbir Singh66e17072008-02-07 00:13:56 -0800988}
989
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800990void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
991{
992 struct page_cgroup *pc;
993 struct mem_cgroup_per_zone *mz;
994
Hirokazu Takahashif8d66542009-01-07 18:08:02 -0800995 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800996 return;
997 pc = lookup_page_cgroup(page);
Balbir Singh4b3bde42009-09-23 15:56:32 -0700998 VM_BUG_ON(PageCgroupAcctLRU(pc));
Johannes Weinera61ed3c2011-11-02 13:38:29 -0700999 /*
1000 * putback: charge:
1001 * SetPageLRU SetPageCgroupUsed
1002 * smp_mb smp_mb
1003 * PageCgroupUsed && add to memcg LRU PageLRU && add to memcg LRU
1004 *
1005 * Ensure that one of the two sides adds the page to the memcg
1006 * LRU during a race.
1007 */
1008 smp_mb();
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001009 if (!PageCgroupUsed(pc))
1010 return;
Johannes Weiner713735b2011-01-20 14:44:31 -08001011 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1012 smp_rmb();
Johannes Weiner97a6c372011-03-23 16:42:27 -07001013 mz = page_cgroup_zoneinfo(pc->mem_cgroup, page);
KAMEZAWA Hiroyukiece35ca2011-01-20 14:44:24 -08001014 /* huge page split is done under lru_lock. so, we have no races. */
1015 MEM_CGROUP_ZSTAT(mz, lru) += 1 << compound_order(page);
Balbir Singh4b3bde42009-09-23 15:56:32 -07001016 SetPageCgroupAcctLRU(pc);
1017 if (mem_cgroup_is_root(pc->mem_cgroup))
1018 return;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001019 list_add(&pc->lru, &mz->lists[lru]);
1020}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001021
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001022/*
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07001023 * At handling SwapCache and other FUSE stuff, pc->mem_cgroup may be changed
1024 * while it's linked to lru because the page may be reused after it's fully
1025 * uncharged. To handle that, unlink page_cgroup from LRU when charge it again.
1026 * It's done under lock_page and expected that zone->lru_lock isnever held.
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001027 */
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07001028static void mem_cgroup_lru_del_before_commit(struct page *page)
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001029{
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001030 unsigned long flags;
1031 struct zone *zone = page_zone(page);
1032 struct page_cgroup *pc = lookup_page_cgroup(page);
1033
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07001034 /*
1035 * Doing this check without taking ->lru_lock seems wrong but this
1036 * is safe. Because if page_cgroup's USED bit is unset, the page
1037 * will not be added to any memcg's LRU. If page_cgroup's USED bit is
1038 * set, the commit after this will fail, anyway.
1039 * This all charge/uncharge is done under some mutual execustion.
1040 * So, we don't need to taking care of changes in USED bit.
1041 */
1042 if (likely(!PageLRU(page)))
1043 return;
1044
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001045 spin_lock_irqsave(&zone->lru_lock, flags);
1046 /*
1047 * Forget old LRU when this page_cgroup is *not* used. This Used bit
1048 * is guarded by lock_page() because the page is SwapCache.
1049 */
1050 if (!PageCgroupUsed(pc))
1051 mem_cgroup_del_lru_list(page, page_lru(page));
1052 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001053}
1054
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07001055static void mem_cgroup_lru_add_after_commit(struct page *page)
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001056{
1057 unsigned long flags;
1058 struct zone *zone = page_zone(page);
1059 struct page_cgroup *pc = lookup_page_cgroup(page);
Johannes Weinera61ed3c2011-11-02 13:38:29 -07001060 /*
1061 * putback: charge:
1062 * SetPageLRU SetPageCgroupUsed
1063 * smp_mb smp_mb
1064 * PageCgroupUsed && add to memcg LRU PageLRU && add to memcg LRU
1065 *
1066 * Ensure that one of the two sides adds the page to the memcg
1067 * LRU during a race.
1068 */
1069 smp_mb();
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07001070 /* taking care of that the page is added to LRU while we commit it */
1071 if (likely(!PageLRU(page)))
1072 return;
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001073 spin_lock_irqsave(&zone->lru_lock, flags);
1074 /* link when the page is linked to LRU but page_cgroup isn't */
Balbir Singh4b3bde42009-09-23 15:56:32 -07001075 if (PageLRU(page) && !PageCgroupAcctLRU(pc))
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001076 mem_cgroup_add_lru_list(page, page_lru(page));
1077 spin_unlock_irqrestore(&zone->lru_lock, flags);
1078}
1079
1080
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001081void mem_cgroup_move_lists(struct page *page,
1082 enum lru_list from, enum lru_list to)
1083{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08001084 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001085 return;
1086 mem_cgroup_del_lru_list(page, from);
1087 mem_cgroup_add_lru_list(page, to);
1088}
1089
Michal Hocko3e920412011-07-26 16:08:29 -07001090/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001091 * Checks whether given mem is same or in the root_mem_cgroup's
Michal Hocko3e920412011-07-26 16:08:29 -07001092 * hierarchy subtree
1093 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001094static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1095 struct mem_cgroup *memcg)
Michal Hocko3e920412011-07-26 16:08:29 -07001096{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001097 if (root_memcg != memcg) {
1098 return (root_memcg->use_hierarchy &&
1099 css_is_ancestor(&memcg->css, &root_memcg->css));
Michal Hocko3e920412011-07-26 16:08:29 -07001100 }
1101
1102 return true;
1103}
1104
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001105int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *memcg)
David Rientjes4c4a2212008-02-07 00:14:06 -08001106{
1107 int ret;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001108 struct mem_cgroup *curr = NULL;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001109 struct task_struct *p;
David Rientjes4c4a2212008-02-07 00:14:06 -08001110
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001111 p = find_lock_task_mm(task);
1112 if (!p)
1113 return 0;
1114 curr = try_get_mem_cgroup_from_mm(p->mm);
1115 task_unlock(p);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001116 if (!curr)
1117 return 0;
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001118 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001119 * We should check use_hierarchy of "memcg" not "curr". Because checking
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001120 * use_hierarchy of "curr" here make this function true if hierarchy is
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001121 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1122 * hierarchy(even if use_hierarchy is disabled in "memcg").
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001123 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001124 ret = mem_cgroup_same_or_subtree(memcg, curr);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001125 css_put(&curr->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001126 return ret;
1127}
1128
Johannes Weiner9b272972011-11-02 13:38:23 -07001129int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg, struct zone *zone)
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001130{
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001131 unsigned long inactive_ratio;
Johannes Weiner9b272972011-11-02 13:38:23 -07001132 int nid = zone_to_nid(zone);
1133 int zid = zone_idx(zone);
1134 unsigned long inactive;
1135 unsigned long active;
1136 unsigned long gb;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001137
Johannes Weiner9b272972011-11-02 13:38:23 -07001138 inactive = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid,
1139 BIT(LRU_INACTIVE_ANON));
1140 active = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid,
1141 BIT(LRU_ACTIVE_ANON));
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001142
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001143 gb = (inactive + active) >> (30 - PAGE_SHIFT);
1144 if (gb)
1145 inactive_ratio = int_sqrt(10 * gb);
1146 else
1147 inactive_ratio = 1;
1148
Johannes Weiner9b272972011-11-02 13:38:23 -07001149 return inactive * inactive_ratio < active;
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001150}
1151
Johannes Weiner9b272972011-11-02 13:38:23 -07001152int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg, struct zone *zone)
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001153{
1154 unsigned long active;
1155 unsigned long inactive;
Johannes Weiner9b272972011-11-02 13:38:23 -07001156 int zid = zone_idx(zone);
1157 int nid = zone_to_nid(zone);
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001158
Johannes Weiner9b272972011-11-02 13:38:23 -07001159 inactive = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid,
1160 BIT(LRU_INACTIVE_FILE));
1161 active = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid,
1162 BIT(LRU_ACTIVE_FILE));
Rik van Riel56e49d22009-06-16 15:32:28 -07001163
1164 return (active > inactive);
1165}
1166
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -08001167struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
1168 struct zone *zone)
1169{
KOSAKI Motohiro13d7e3a2010-08-10 18:03:06 -07001170 int nid = zone_to_nid(zone);
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -08001171 int zid = zone_idx(zone);
1172 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
1173
1174 return &mz->reclaim_stat;
1175}
1176
1177struct zone_reclaim_stat *
1178mem_cgroup_get_reclaim_stat_from_page(struct page *page)
1179{
1180 struct page_cgroup *pc;
1181 struct mem_cgroup_per_zone *mz;
1182
1183 if (mem_cgroup_disabled())
1184 return NULL;
1185
1186 pc = lookup_page_cgroup(page);
Daisuke Nishimurabd112db2009-01-15 13:51:11 -08001187 if (!PageCgroupUsed(pc))
1188 return NULL;
Johannes Weiner713735b2011-01-20 14:44:31 -08001189 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1190 smp_rmb();
Johannes Weiner97a6c372011-03-23 16:42:27 -07001191 mz = page_cgroup_zoneinfo(pc->mem_cgroup, page);
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -08001192 return &mz->reclaim_stat;
1193}
1194
Balbir Singh66e17072008-02-07 00:13:56 -08001195unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
1196 struct list_head *dst,
1197 unsigned long *scanned, int order,
Minchan Kim4356f212011-10-31 17:06:47 -07001198 isolate_mode_t mode,
1199 struct zone *z,
Balbir Singh66e17072008-02-07 00:13:56 -08001200 struct mem_cgroup *mem_cont,
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001201 int active, int file)
Balbir Singh66e17072008-02-07 00:13:56 -08001202{
1203 unsigned long nr_taken = 0;
1204 struct page *page;
1205 unsigned long scan;
1206 LIST_HEAD(pc_list);
1207 struct list_head *src;
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -08001208 struct page_cgroup *pc, *tmp;
KOSAKI Motohiro13d7e3a2010-08-10 18:03:06 -07001209 int nid = zone_to_nid(z);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08001210 int zid = zone_idx(z);
1211 struct mem_cgroup_per_zone *mz;
Johannes Weinerb7c46d12009-09-21 17:02:56 -07001212 int lru = LRU_FILE * file + active;
KAMEZAWA Hiroyuki2ffebca2009-06-17 16:27:21 -07001213 int ret;
Balbir Singh66e17072008-02-07 00:13:56 -08001214
Balbir Singhcf475ad2008-04-29 01:00:16 -07001215 BUG_ON(!mem_cont);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08001216 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
Christoph Lameterb69408e2008-10-18 20:26:14 -07001217 src = &mz->lists[lru];
Balbir Singh66e17072008-02-07 00:13:56 -08001218
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -08001219 scan = 0;
1220 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
Hugh Dickins436c65412008-02-07 00:14:12 -08001221 if (scan >= nr_to_scan)
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -08001222 break;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001223
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001224 if (unlikely(!PageCgroupUsed(pc)))
1225 continue;
Johannes Weiner5564e882011-03-23 16:42:29 -07001226
Johannes Weiner6b3ae582011-03-23 16:42:30 -07001227 page = lookup_cgroup_page(pc);
Johannes Weiner5564e882011-03-23 16:42:29 -07001228
Hugh Dickins436c65412008-02-07 00:14:12 -08001229 if (unlikely(!PageLRU(page)))
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -08001230 continue;
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -08001231
Hugh Dickins436c65412008-02-07 00:14:12 -08001232 scan++;
KAMEZAWA Hiroyuki2ffebca2009-06-17 16:27:21 -07001233 ret = __isolate_lru_page(page, mode, file);
1234 switch (ret) {
1235 case 0:
Balbir Singh66e17072008-02-07 00:13:56 -08001236 list_move(&page->lru, dst);
KAMEZAWA Hiroyuki2ffebca2009-06-17 16:27:21 -07001237 mem_cgroup_del_lru(page);
Rik van Riel2c888cf2011-01-13 15:47:13 -08001238 nr_taken += hpage_nr_pages(page);
KAMEZAWA Hiroyuki2ffebca2009-06-17 16:27:21 -07001239 break;
1240 case -EBUSY:
1241 /* we don't affect global LRU but rotate in our LRU */
1242 mem_cgroup_rotate_lru_list(page, page_lru(page));
1243 break;
1244 default:
1245 break;
Balbir Singh66e17072008-02-07 00:13:56 -08001246 }
1247 }
1248
Balbir Singh66e17072008-02-07 00:13:56 -08001249 *scanned = scan;
KOSAKI Motohirocc8e9702010-08-09 17:19:57 -07001250
1251 trace_mm_vmscan_memcg_isolate(0, nr_to_scan, scan, nr_taken,
1252 0, 0, 0, mode);
1253
Balbir Singh66e17072008-02-07 00:13:56 -08001254 return nr_taken;
1255}
1256
Balbir Singh6d61ef42009-01-07 18:08:06 -08001257#define mem_cgroup_from_res_counter(counter, member) \
1258 container_of(counter, struct mem_cgroup, member)
1259
Johannes Weiner19942822011-02-01 15:52:43 -08001260/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001261 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1262 * @mem: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001263 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001264 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001265 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001266 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001267static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001268{
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001269 unsigned long long margin;
1270
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001271 margin = res_counter_margin(&memcg->res);
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001272 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001273 margin = min(margin, res_counter_margin(&memcg->memsw));
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001274 return margin >> PAGE_SHIFT;
Johannes Weiner19942822011-02-01 15:52:43 -08001275}
1276
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07001277int mem_cgroup_swappiness(struct mem_cgroup *memcg)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001278{
1279 struct cgroup *cgrp = memcg->css.cgroup;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001280
1281 /* root ? */
1282 if (cgrp->parent == NULL)
1283 return vm_swappiness;
1284
Johannes Weinerbf1ff262011-03-23 16:42:32 -07001285 return memcg->swappiness;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001286}
1287
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001288static void mem_cgroup_start_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001289{
1290 int cpu;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001291
1292 get_online_cpus();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001293 spin_lock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001294 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001295 per_cpu(memcg->stat->count[MEM_CGROUP_ON_MOVE], cpu) += 1;
1296 memcg->nocpu_base.count[MEM_CGROUP_ON_MOVE] += 1;
1297 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001298 put_online_cpus();
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001299
1300 synchronize_rcu();
1301}
1302
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001303static void mem_cgroup_end_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001304{
1305 int cpu;
1306
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001307 if (!memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001308 return;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001309 get_online_cpus();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001310 spin_lock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001311 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001312 per_cpu(memcg->stat->count[MEM_CGROUP_ON_MOVE], cpu) -= 1;
1313 memcg->nocpu_base.count[MEM_CGROUP_ON_MOVE] -= 1;
1314 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001315 put_online_cpus();
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001316}
1317/*
1318 * 2 routines for checking "mem" is under move_account() or not.
1319 *
1320 * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used
1321 * for avoiding race in accounting. If true,
1322 * pc->mem_cgroup may be overwritten.
1323 *
1324 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1325 * under hierarchy of moving cgroups. This is for
1326 * waiting at hith-memory prressure caused by "move".
1327 */
1328
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001329static bool mem_cgroup_stealed(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001330{
1331 VM_BUG_ON(!rcu_read_lock_held());
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001332 return this_cpu_read(memcg->stat->count[MEM_CGROUP_ON_MOVE]) > 0;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001333}
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001334
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001335static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001336{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001337 struct mem_cgroup *from;
1338 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001339 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001340 /*
1341 * Unlike task_move routines, we access mc.to, mc.from not under
1342 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1343 */
1344 spin_lock(&mc.lock);
1345 from = mc.from;
1346 to = mc.to;
1347 if (!from)
1348 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001349
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001350 ret = mem_cgroup_same_or_subtree(memcg, from)
1351 || mem_cgroup_same_or_subtree(memcg, to);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001352unlock:
1353 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001354 return ret;
1355}
1356
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001357static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001358{
1359 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001360 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001361 DEFINE_WAIT(wait);
1362 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1363 /* moving charge context might have finished. */
1364 if (mc.moving_task)
1365 schedule();
1366 finish_wait(&mc.waitq, &wait);
1367 return true;
1368 }
1369 }
1370 return false;
1371}
1372
Balbir Singhe2224322009-04-02 16:57:39 -07001373/**
Kirill A. Shutemov6a6135b2010-03-10 15:22:25 -08001374 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
Balbir Singhe2224322009-04-02 16:57:39 -07001375 * @memcg: The memory cgroup that went over limit
1376 * @p: Task that is going to be killed
1377 *
1378 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1379 * enabled
1380 */
1381void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1382{
1383 struct cgroup *task_cgrp;
1384 struct cgroup *mem_cgrp;
1385 /*
1386 * Need a buffer in BSS, can't rely on allocations. The code relies
1387 * on the assumption that OOM is serialized for memory controller.
1388 * If this assumption is broken, revisit this code.
1389 */
1390 static char memcg_name[PATH_MAX];
1391 int ret;
1392
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001393 if (!memcg || !p)
Balbir Singhe2224322009-04-02 16:57:39 -07001394 return;
1395
1396
1397 rcu_read_lock();
1398
1399 mem_cgrp = memcg->css.cgroup;
1400 task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
1401
1402 ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
1403 if (ret < 0) {
1404 /*
1405 * Unfortunately, we are unable to convert to a useful name
1406 * But we'll still print out the usage information
1407 */
1408 rcu_read_unlock();
1409 goto done;
1410 }
1411 rcu_read_unlock();
1412
1413 printk(KERN_INFO "Task in %s killed", memcg_name);
1414
1415 rcu_read_lock();
1416 ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
1417 if (ret < 0) {
1418 rcu_read_unlock();
1419 goto done;
1420 }
1421 rcu_read_unlock();
1422
1423 /*
1424 * Continues from above, so we don't need an KERN_ level
1425 */
1426 printk(KERN_CONT " as a result of limit of %s\n", memcg_name);
1427done:
1428
1429 printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n",
1430 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1431 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1432 res_counter_read_u64(&memcg->res, RES_FAILCNT));
1433 printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, "
1434 "failcnt %llu\n",
1435 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1436 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1437 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
1438}
1439
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001440/*
1441 * This function returns the number of memcg under hierarchy tree. Returns
1442 * 1(self count) if no children.
1443 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001444static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001445{
1446 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001447 struct mem_cgroup *iter;
1448
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001449 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001450 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001451 return num;
1452}
1453
Balbir Singh6d61ef42009-01-07 18:08:06 -08001454/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001455 * Return the memory (and swap, if configured) limit for a memcg.
1456 */
1457u64 mem_cgroup_get_limit(struct mem_cgroup *memcg)
1458{
1459 u64 limit;
1460 u64 memsw;
1461
Johannes Weinerf3e8eb72011-01-13 15:47:39 -08001462 limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1463 limit += total_swap_pages << PAGE_SHIFT;
1464
David Rientjesa63d83f2010-08-09 17:19:46 -07001465 memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1466 /*
1467 * If memsw is finite and limits the amount of swap space available
1468 * to this memcg, return that limit.
1469 */
1470 return min(limit, memsw);
1471}
1472
1473/*
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001474 * Visit the first child (need not be the first child as per the ordering
1475 * of the cgroup list, since we track last_scanned_child) of @mem and use
1476 * that to reclaim free pages from.
1477 */
1478static struct mem_cgroup *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001479mem_cgroup_select_victim(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001480{
1481 struct mem_cgroup *ret = NULL;
1482 struct cgroup_subsys_state *css;
1483 int nextid, found;
1484
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001485 if (!root_memcg->use_hierarchy) {
1486 css_get(&root_memcg->css);
1487 ret = root_memcg;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001488 }
1489
1490 while (!ret) {
1491 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001492 nextid = root_memcg->last_scanned_child + 1;
1493 css = css_get_next(&mem_cgroup_subsys, nextid, &root_memcg->css,
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001494 &found);
1495 if (css && css_tryget(css))
1496 ret = container_of(css, struct mem_cgroup, css);
1497
1498 rcu_read_unlock();
1499 /* Updates scanning parameter */
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001500 if (!css) {
1501 /* this means start scan from ID:1 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001502 root_memcg->last_scanned_child = 0;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001503 } else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001504 root_memcg->last_scanned_child = found;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001505 }
1506
1507 return ret;
1508}
1509
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001510/**
1511 * test_mem_cgroup_node_reclaimable
1512 * @mem: the target memcg
1513 * @nid: the node ID to be checked.
1514 * @noswap : specify true here if the user wants flle only information.
1515 *
1516 * This function returns whether the specified memcg contains any
1517 * reclaimable pages on a node. Returns true if there are any reclaimable
1518 * pages in the node.
1519 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001520static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001521 int nid, bool noswap)
1522{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001523 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001524 return true;
1525 if (noswap || !total_swap_pages)
1526 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001527 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001528 return true;
1529 return false;
1530
1531}
Ying Han889976d2011-05-26 16:25:33 -07001532#if MAX_NUMNODES > 1
1533
1534/*
1535 * Always updating the nodemask is not very good - even if we have an empty
1536 * list or the wrong list here, we can start from some node and traverse all
1537 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1538 *
1539 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001540static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001541{
1542 int nid;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001543 /*
1544 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1545 * pagein/pageout changes since the last update.
1546 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001547 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001548 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001549 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001550 return;
1551
Ying Han889976d2011-05-26 16:25:33 -07001552 /* make a nodemask where this memcg uses memory from */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001553 memcg->scan_nodes = node_states[N_HIGH_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001554
1555 for_each_node_mask(nid, node_states[N_HIGH_MEMORY]) {
1556
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001557 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1558 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001559 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001560
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001561 atomic_set(&memcg->numainfo_events, 0);
1562 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001563}
1564
1565/*
1566 * Selecting a node where we start reclaim from. Because what we need is just
1567 * reducing usage counter, start from anywhere is O,K. Considering
1568 * memory reclaim from current node, there are pros. and cons.
1569 *
1570 * Freeing memory from current node means freeing memory from a node which
1571 * we'll use or we've used. So, it may make LRU bad. And if several threads
1572 * hit limits, it will see a contention on a node. But freeing from remote
1573 * node means more costs for memory reclaim because of memory latency.
1574 *
1575 * Now, we use round-robin. Better algorithm is welcomed.
1576 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001577int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001578{
1579 int node;
1580
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001581 mem_cgroup_may_update_nodemask(memcg);
1582 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001583
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001584 node = next_node(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001585 if (node == MAX_NUMNODES)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001586 node = first_node(memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001587 /*
1588 * We call this when we hit limit, not when pages are added to LRU.
1589 * No LRU may hold pages because all pages are UNEVICTABLE or
1590 * memcg is too small and all pages are not on LRU. In that case,
1591 * we use curret node.
1592 */
1593 if (unlikely(node == MAX_NUMNODES))
1594 node = numa_node_id();
1595
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001596 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001597 return node;
1598}
1599
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001600/*
1601 * Check all nodes whether it contains reclaimable pages or not.
1602 * For quick scan, we make use of scan_nodes. This will allow us to skip
1603 * unused nodes. But scan_nodes is lazily updated and may not cotain
1604 * enough new information. We need to do double check.
1605 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001606bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001607{
1608 int nid;
1609
1610 /*
1611 * quick check...making use of scan_node.
1612 * We can skip unused nodes.
1613 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001614 if (!nodes_empty(memcg->scan_nodes)) {
1615 for (nid = first_node(memcg->scan_nodes);
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001616 nid < MAX_NUMNODES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001617 nid = next_node(nid, memcg->scan_nodes)) {
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001618
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001619 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001620 return true;
1621 }
1622 }
1623 /*
1624 * Check rest of nodes.
1625 */
1626 for_each_node_state(nid, N_HIGH_MEMORY) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001627 if (node_isset(nid, memcg->scan_nodes))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001628 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001629 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001630 return true;
1631 }
1632 return false;
1633}
1634
Ying Han889976d2011-05-26 16:25:33 -07001635#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001636int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001637{
1638 return 0;
1639}
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001640
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001641bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001642{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001643 return test_mem_cgroup_node_reclaimable(memcg, 0, noswap);
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001644}
Ying Han889976d2011-05-26 16:25:33 -07001645#endif
1646
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001647/*
1648 * Scan the hierarchy if needed to reclaim memory. We remember the last child
1649 * we reclaimed from, so that we don't end up penalizing one child extensively
1650 * based on its position in the children list.
Balbir Singh6d61ef42009-01-07 18:08:06 -08001651 *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001652 * root_memcg is the original ancestor that we've been reclaim from.
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001653 *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001654 * We give up and return to the caller when we visit root_memcg twice.
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001655 * (other groups can be removed while we're walking....)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001656 *
1657 * If shrink==true, for avoiding to free too much, this returns immedieately.
Balbir Singh6d61ef42009-01-07 18:08:06 -08001658 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001659static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_memcg,
Balbir Singh4e416952009-09-23 15:56:39 -07001660 struct zone *zone,
Balbir Singh75822b42009-09-23 15:56:38 -07001661 gfp_t gfp_mask,
Ying Han0ae5e892011-05-26 16:25:25 -07001662 unsigned long reclaim_options,
1663 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001664{
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001665 struct mem_cgroup *victim;
1666 int ret, total = 0;
1667 int loop = 0;
Balbir Singh75822b42009-09-23 15:56:38 -07001668 bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP;
1669 bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK;
Balbir Singh4e416952009-09-23 15:56:39 -07001670 bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001671 unsigned long excess;
Johannes Weiner185efc02011-09-14 16:21:58 -07001672 unsigned long nr_scanned;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001673
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001674 excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001675
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07001676 /* If memsw_is_minimum==1, swap-out is of-no-use. */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001677 if (!check_soft && !shrink && root_memcg->memsw_is_minimum)
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07001678 noswap = true;
1679
Balbir Singh4e416952009-09-23 15:56:39 -07001680 while (1) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001681 victim = mem_cgroup_select_victim(root_memcg);
1682 if (victim == root_memcg) {
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001683 loop++;
KAMEZAWA Hiroyukifbc29a22011-06-15 15:08:46 -07001684 /*
1685 * We are not draining per cpu cached charges during
1686 * soft limit reclaim because global reclaim doesn't
1687 * care about charges. It tries to free some memory and
1688 * charges will not give any.
1689 */
1690 if (!check_soft && loop >= 1)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001691 drain_all_stock_async(root_memcg);
Balbir Singh4e416952009-09-23 15:56:39 -07001692 if (loop >= 2) {
1693 /*
1694 * If we have not been able to reclaim
1695 * anything, it might because there are
1696 * no reclaimable pages under this hierarchy
1697 */
1698 if (!check_soft || !total) {
1699 css_put(&victim->css);
1700 break;
1701 }
1702 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001703 * We want to do more targeted reclaim.
Balbir Singh4e416952009-09-23 15:56:39 -07001704 * excess >> 2 is not to excessive so as to
1705 * reclaim too much, nor too less that we keep
1706 * coming back to reclaim from this cgroup
1707 */
1708 if (total >= (excess >> 2) ||
1709 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) {
1710 css_put(&victim->css);
1711 break;
1712 }
1713 }
1714 }
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001715 if (!mem_cgroup_reclaimable(victim, noswap)) {
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001716 /* this cgroup's local usage == 0 */
1717 css_put(&victim->css);
Balbir Singh6d61ef42009-01-07 18:08:06 -08001718 continue;
1719 }
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001720 /* we use swappiness of local cgroup */
Ying Han0ae5e892011-05-26 16:25:25 -07001721 if (check_soft) {
Balbir Singh4e416952009-09-23 15:56:39 -07001722 ret = mem_cgroup_shrink_node_zone(victim, gfp_mask,
Johannes Weiner185efc02011-09-14 16:21:58 -07001723 noswap, zone, &nr_scanned);
1724 *total_scanned += nr_scanned;
Ying Han0ae5e892011-05-26 16:25:25 -07001725 } else
Balbir Singh4e416952009-09-23 15:56:39 -07001726 ret = try_to_free_mem_cgroup_pages(victim, gfp_mask,
Johannes Weiner185efc02011-09-14 16:21:58 -07001727 noswap);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001728 css_put(&victim->css);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001729 /*
1730 * At shrinking usage, we can't check we should stop here or
1731 * reclaim more. It's depends on callers. last_scanned_child
1732 * will work enough for keeping fairness under tree.
1733 */
1734 if (shrink)
1735 return ret;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001736 total += ret;
Balbir Singh4e416952009-09-23 15:56:39 -07001737 if (check_soft) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001738 if (!res_counter_soft_limit_excess(&root_memcg->res))
Balbir Singh4e416952009-09-23 15:56:39 -07001739 return total;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001740 } else if (mem_cgroup_margin(root_memcg))
Johannes Weiner4fd14eb2011-05-26 16:25:35 -07001741 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001742 }
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001743 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001744}
1745
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001746/*
1747 * Check OOM-Killer is already running under our hierarchy.
1748 * If someone is running, return false.
Michal Hocko1af8efe2011-07-26 16:08:24 -07001749 * Has to be called with memcg_oom_lock
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001750 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001751static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001752{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001753 struct mem_cgroup *iter, *failed = NULL;
1754 bool cond = true;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001755
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001756 for_each_mem_cgroup_tree_cond(iter, memcg, cond) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001757 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001758 /*
1759 * this subtree of our hierarchy is already locked
1760 * so we cannot give a lock.
1761 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001762 failed = iter;
1763 cond = false;
Johannes Weiner23751be2011-08-25 15:59:16 -07001764 } else
1765 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001766 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001767
Michal Hocko79dfdac2011-07-26 16:08:23 -07001768 if (!failed)
Johannes Weiner23751be2011-08-25 15:59:16 -07001769 return true;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001770
1771 /*
1772 * OK, we failed to lock the whole subtree so we have to clean up
1773 * what we set up to the failing subtree
1774 */
1775 cond = true;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001776 for_each_mem_cgroup_tree_cond(iter, memcg, cond) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001777 if (iter == failed) {
1778 cond = false;
1779 continue;
1780 }
1781 iter->oom_lock = false;
1782 }
Johannes Weiner23751be2011-08-25 15:59:16 -07001783 return false;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001784}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001785
Michal Hocko79dfdac2011-07-26 16:08:23 -07001786/*
Michal Hocko1af8efe2011-07-26 16:08:24 -07001787 * Has to be called with memcg_oom_lock
Michal Hocko79dfdac2011-07-26 16:08:23 -07001788 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001789static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001790{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001791 struct mem_cgroup *iter;
1792
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001793 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001794 iter->oom_lock = false;
1795 return 0;
1796}
1797
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001798static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001799{
1800 struct mem_cgroup *iter;
1801
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001802 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001803 atomic_inc(&iter->under_oom);
1804}
1805
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001806static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001807{
1808 struct mem_cgroup *iter;
1809
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001810 /*
1811 * When a new child is created while the hierarchy is under oom,
1812 * mem_cgroup_oom_lock() may not be called. We have to use
1813 * atomic_add_unless() here.
1814 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001815 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001816 atomic_add_unless(&iter->under_oom, -1, 0);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001817}
1818
Michal Hocko1af8efe2011-07-26 16:08:24 -07001819static DEFINE_SPINLOCK(memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001820static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1821
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001822struct oom_wait_info {
1823 struct mem_cgroup *mem;
1824 wait_queue_t wait;
1825};
1826
1827static int memcg_oom_wake_function(wait_queue_t *wait,
1828 unsigned mode, int sync, void *arg)
1829{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001830 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg,
1831 *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001832 struct oom_wait_info *oom_wait_info;
1833
1834 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001835 oom_wait_memcg = oom_wait_info->mem;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001836
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001837 /*
1838 * Both of oom_wait_info->mem and wake_mem are stable under us.
1839 * Then we can use css_is_ancestor without taking care of RCU.
1840 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001841 if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg)
1842 && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001843 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001844 return autoremove_wake_function(wait, mode, sync, arg);
1845}
1846
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001847static void memcg_wakeup_oom(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001848{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001849 /* for filtering, pass "memcg" as argument. */
1850 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001851}
1852
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001853static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001854{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001855 if (memcg && atomic_read(&memcg->under_oom))
1856 memcg_wakeup_oom(memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001857}
1858
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001859/*
1860 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1861 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001862bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001863{
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001864 struct oom_wait_info owait;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001865 bool locked, need_to_kill;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001866
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001867 owait.mem = memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001868 owait.wait.flags = 0;
1869 owait.wait.func = memcg_oom_wake_function;
1870 owait.wait.private = current;
1871 INIT_LIST_HEAD(&owait.wait.task_list);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001872 need_to_kill = true;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001873 mem_cgroup_mark_under_oom(memcg);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001874
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001875 /* At first, try to OOM lock hierarchy under memcg.*/
Michal Hocko1af8efe2011-07-26 16:08:24 -07001876 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001877 locked = mem_cgroup_oom_lock(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001878 /*
1879 * Even if signal_pending(), we can't quit charge() loop without
1880 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1881 * under OOM is always welcomed, use TASK_KILLABLE here.
1882 */
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001883 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001884 if (!locked || memcg->oom_kill_disable)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001885 need_to_kill = false;
1886 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001887 mem_cgroup_oom_notify(memcg);
Michal Hocko1af8efe2011-07-26 16:08:24 -07001888 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001889
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001890 if (need_to_kill) {
1891 finish_wait(&memcg_oom_waitq, &owait.wait);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001892 mem_cgroup_out_of_memory(memcg, mask);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001893 } else {
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001894 schedule();
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001895 finish_wait(&memcg_oom_waitq, &owait.wait);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001896 }
Michal Hocko1af8efe2011-07-26 16:08:24 -07001897 spin_lock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001898 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001899 mem_cgroup_oom_unlock(memcg);
1900 memcg_wakeup_oom(memcg);
Michal Hocko1af8efe2011-07-26 16:08:24 -07001901 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001902
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001903 mem_cgroup_unmark_under_oom(memcg);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001904
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001905 if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current))
1906 return false;
1907 /* Give chance to dying process */
KAMEZAWA Hiroyuki715a5ee2011-11-02 13:38:18 -07001908 schedule_timeout_uninterruptible(1);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001909 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001910}
1911
Balbir Singhd69b0422009-06-17 16:26:34 -07001912/*
1913 * Currently used to update mapped file statistics, but the routine can be
1914 * generalized to update other statistics as well.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001915 *
1916 * Notes: Race condition
1917 *
1918 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1919 * it tends to be costly. But considering some conditions, we doesn't need
1920 * to do so _always_.
1921 *
1922 * Considering "charge", lock_page_cgroup() is not required because all
1923 * file-stat operations happen after a page is attached to radix-tree. There
1924 * are no race with "charge".
1925 *
1926 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1927 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1928 * if there are race with "uncharge". Statistics itself is properly handled
1929 * by flags.
1930 *
1931 * Considering "move", this is an only case we see a race. To make the race
1932 * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are
1933 * possibility of race condition. If there is, we take a lock.
Balbir Singhd69b0422009-06-17 16:26:34 -07001934 */
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001935
Greg Thelen2a7106f2011-01-13 15:47:37 -08001936void mem_cgroup_update_page_stat(struct page *page,
1937 enum mem_cgroup_page_stat_item idx, int val)
Balbir Singhd69b0422009-06-17 16:26:34 -07001938{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001939 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001940 struct page_cgroup *pc = lookup_page_cgroup(page);
1941 bool need_unlock = false;
KAMEZAWA Hiroyukidbd4ea72011-01-13 15:47:38 -08001942 unsigned long uninitialized_var(flags);
Balbir Singhd69b0422009-06-17 16:26:34 -07001943
Balbir Singhd69b0422009-06-17 16:26:34 -07001944 if (unlikely(!pc))
1945 return;
1946
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001947 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001948 memcg = pc->mem_cgroup;
1949 if (unlikely(!memcg || !PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001950 goto out;
1951 /* pc->mem_cgroup is unstable ? */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001952 if (unlikely(mem_cgroup_stealed(memcg)) || PageTransHuge(page)) {
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001953 /* take a lock against to access pc->mem_cgroup */
KAMEZAWA Hiroyukidbd4ea72011-01-13 15:47:38 -08001954 move_lock_page_cgroup(pc, &flags);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001955 need_unlock = true;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001956 memcg = pc->mem_cgroup;
1957 if (!memcg || !PageCgroupUsed(pc))
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001958 goto out;
1959 }
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001960
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001961 switch (idx) {
Greg Thelen2a7106f2011-01-13 15:47:37 -08001962 case MEMCG_NR_FILE_MAPPED:
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001963 if (val > 0)
1964 SetPageCgroupFileMapped(pc);
1965 else if (!page_mapped(page))
KAMEZAWA Hiroyuki0c270f82010-10-27 15:33:39 -07001966 ClearPageCgroupFileMapped(pc);
Greg Thelen2a7106f2011-01-13 15:47:37 -08001967 idx = MEM_CGROUP_STAT_FILE_MAPPED;
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001968 break;
1969 default:
1970 BUG();
KAMEZAWA Hiroyuki8725d542010-04-06 14:35:05 -07001971 }
Balbir Singhd69b0422009-06-17 16:26:34 -07001972
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001973 this_cpu_add(memcg->stat->count[idx], val);
Greg Thelen2a7106f2011-01-13 15:47:37 -08001974
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001975out:
1976 if (unlikely(need_unlock))
KAMEZAWA Hiroyukidbd4ea72011-01-13 15:47:38 -08001977 move_unlock_page_cgroup(pc, &flags);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001978 rcu_read_unlock();
1979 return;
Balbir Singhd69b0422009-06-17 16:26:34 -07001980}
Greg Thelen2a7106f2011-01-13 15:47:37 -08001981EXPORT_SYMBOL(mem_cgroup_update_page_stat);
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001982
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001983/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001984 * size of first charge trial. "32" comes from vmscan.c's magic value.
1985 * TODO: maybe necessary to use big numbers in big irons.
1986 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001987#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001988struct memcg_stock_pcp {
1989 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001990 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001991 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001992 unsigned long flags;
1993#define FLUSHING_CACHED_CHARGE (0)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001994};
1995static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02001996static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001997
1998/*
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001999 * Try to consume stocked charge on this cpu. If success, one page is consumed
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002000 * from local stock and true is returned. If the stock is 0 or charges from a
2001 * cgroup which is not current target, returns false. This stock will be
2002 * refilled.
2003 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002004static bool consume_stock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002005{
2006 struct memcg_stock_pcp *stock;
2007 bool ret = true;
2008
2009 stock = &get_cpu_var(memcg_stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002010 if (memcg == stock->cached && stock->nr_pages)
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002011 stock->nr_pages--;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002012 else /* need to call res_counter_charge */
2013 ret = false;
2014 put_cpu_var(memcg_stock);
2015 return ret;
2016}
2017
2018/*
2019 * Returns stocks cached in percpu to res_counter and reset cached information.
2020 */
2021static void drain_stock(struct memcg_stock_pcp *stock)
2022{
2023 struct mem_cgroup *old = stock->cached;
2024
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002025 if (stock->nr_pages) {
2026 unsigned long bytes = stock->nr_pages * PAGE_SIZE;
2027
2028 res_counter_uncharge(&old->res, bytes);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002029 if (do_swap_account)
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002030 res_counter_uncharge(&old->memsw, bytes);
2031 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002032 }
2033 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002034}
2035
2036/*
2037 * This must be called under preempt disabled or must be called by
2038 * a thread which is pinned to local cpu.
2039 */
2040static void drain_local_stock(struct work_struct *dummy)
2041{
2042 struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
2043 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002044 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002045}
2046
2047/*
2048 * Cache charges(val) which is from res_counter, to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01002049 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002050 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002051static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002052{
2053 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
2054
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002055 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002056 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002057 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002058 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002059 stock->nr_pages += nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002060 put_cpu_var(memcg_stock);
2061}
2062
2063/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002064 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Michal Hockod38144b2011-07-26 16:08:28 -07002065 * of the hierarchy under it. sync flag says whether we should block
2066 * until the work is done.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002067 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002068static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002069{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002070 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07002071
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002072 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002073 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07002074 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002075 for_each_online_cpu(cpu) {
2076 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002077 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002078
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002079 memcg = stock->cached;
2080 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002081 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002082 if (!mem_cgroup_same_or_subtree(root_memcg, memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07002083 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07002084 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
2085 if (cpu == curcpu)
2086 drain_local_stock(&stock->work);
2087 else
2088 schedule_work_on(cpu, &stock->work);
2089 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002090 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07002091 put_cpu();
Michal Hockod38144b2011-07-26 16:08:28 -07002092
2093 if (!sync)
2094 goto out;
2095
2096 for_each_online_cpu(cpu) {
2097 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002098 if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags))
Michal Hockod38144b2011-07-26 16:08:28 -07002099 flush_work(&stock->work);
2100 }
2101out:
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002102 put_online_cpus();
Michal Hockod38144b2011-07-26 16:08:28 -07002103}
2104
2105/*
2106 * Tries to drain stocked charges in other cpus. This function is asynchronous
2107 * and just put a work per cpu for draining localy on each cpu. Caller can
2108 * expects some charges will be back to res_counter later but cannot wait for
2109 * it.
2110 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002111static void drain_all_stock_async(struct mem_cgroup *root_memcg)
Michal Hockod38144b2011-07-26 16:08:28 -07002112{
Michal Hocko9f50fad2011-08-09 11:56:26 +02002113 /*
2114 * If someone calls draining, avoid adding more kworker runs.
2115 */
2116 if (!mutex_trylock(&percpu_charge_mutex))
2117 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002118 drain_all_stock(root_memcg, false);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002119 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002120}
2121
2122/* This is a synchronous drain interface. */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002123static void drain_all_stock_sync(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002124{
2125 /* called when force_empty is called */
Michal Hocko9f50fad2011-08-09 11:56:26 +02002126 mutex_lock(&percpu_charge_mutex);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002127 drain_all_stock(root_memcg, true);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002128 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002129}
2130
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002131/*
2132 * This function drains percpu counter value from DEAD cpu and
2133 * move it to local cpu. Note that this function can be preempted.
2134 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002135static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002136{
2137 int i;
2138
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002139 spin_lock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002140 for (i = 0; i < MEM_CGROUP_STAT_DATA; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002141 long x = per_cpu(memcg->stat->count[i], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002142
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002143 per_cpu(memcg->stat->count[i], cpu) = 0;
2144 memcg->nocpu_base.count[i] += x;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002145 }
Johannes Weinere9f89742011-03-23 16:42:37 -07002146 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002147 unsigned long x = per_cpu(memcg->stat->events[i], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -07002148
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002149 per_cpu(memcg->stat->events[i], cpu) = 0;
2150 memcg->nocpu_base.events[i] += x;
Johannes Weinere9f89742011-03-23 16:42:37 -07002151 }
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002152 /* need to clear ON_MOVE value, works as a kind of lock. */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002153 per_cpu(memcg->stat->count[MEM_CGROUP_ON_MOVE], cpu) = 0;
2154 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002155}
2156
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002157static void synchronize_mem_cgroup_on_move(struct mem_cgroup *memcg, int cpu)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002158{
2159 int idx = MEM_CGROUP_ON_MOVE;
2160
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002161 spin_lock(&memcg->pcp_counter_lock);
2162 per_cpu(memcg->stat->count[idx], cpu) = memcg->nocpu_base.count[idx];
2163 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002164}
2165
2166static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002167 unsigned long action,
2168 void *hcpu)
2169{
2170 int cpu = (unsigned long)hcpu;
2171 struct memcg_stock_pcp *stock;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002172 struct mem_cgroup *iter;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002173
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002174 if ((action == CPU_ONLINE)) {
2175 for_each_mem_cgroup_all(iter)
2176 synchronize_mem_cgroup_on_move(iter, cpu);
2177 return NOTIFY_OK;
2178 }
2179
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002180 if ((action != CPU_DEAD) || action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002181 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002182
2183 for_each_mem_cgroup_all(iter)
2184 mem_cgroup_drain_pcp_counter(iter, cpu);
2185
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002186 stock = &per_cpu(memcg_stock, cpu);
2187 drain_stock(stock);
2188 return NOTIFY_OK;
2189}
2190
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002191
2192/* See __mem_cgroup_try_charge() for details */
2193enum {
2194 CHARGE_OK, /* success */
2195 CHARGE_RETRY, /* need to retry but retry is not bad */
2196 CHARGE_NOMEM, /* we can't do more. return -ENOMEM */
2197 CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */
2198 CHARGE_OOM_DIE, /* the current is killed because of OOM */
2199};
2200
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002201static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002202 unsigned int nr_pages, bool oom_check)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002203{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002204 unsigned long csize = nr_pages * PAGE_SIZE;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002205 struct mem_cgroup *mem_over_limit;
2206 struct res_counter *fail_res;
2207 unsigned long flags = 0;
2208 int ret;
2209
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002210 ret = res_counter_charge(&memcg->res, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002211
2212 if (likely(!ret)) {
2213 if (!do_swap_account)
2214 return CHARGE_OK;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002215 ret = res_counter_charge(&memcg->memsw, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002216 if (likely(!ret))
2217 return CHARGE_OK;
2218
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002219 res_counter_uncharge(&memcg->res, csize);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002220 mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw);
2221 flags |= MEM_CGROUP_RECLAIM_NOSWAP;
2222 } else
2223 mem_over_limit = mem_cgroup_from_res_counter(fail_res, res);
Johannes Weiner9221edb2011-02-01 15:52:42 -08002224 /*
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002225 * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2226 * of regular pages (CHARGE_BATCH), or a single regular page (1).
Johannes Weiner9221edb2011-02-01 15:52:42 -08002227 *
2228 * Never reclaim on behalf of optional batching, retry with a
2229 * single page instead.
2230 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002231 if (nr_pages == CHARGE_BATCH)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002232 return CHARGE_RETRY;
2233
2234 if (!(gfp_mask & __GFP_WAIT))
2235 return CHARGE_WOULDBLOCK;
2236
2237 ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL,
Ying Han0ae5e892011-05-26 16:25:25 -07002238 gfp_mask, flags, NULL);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002239 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner19942822011-02-01 15:52:43 -08002240 return CHARGE_RETRY;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002241 /*
Johannes Weiner19942822011-02-01 15:52:43 -08002242 * Even though the limit is exceeded at this point, reclaim
2243 * may have been able to free some pages. Retry the charge
2244 * before killing the task.
2245 *
2246 * Only for regular pages, though: huge pages are rather
2247 * unlikely to succeed so close to the limit, and we fall back
2248 * to regular pages anyway in case of failure.
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002249 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002250 if (nr_pages == 1 && ret)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002251 return CHARGE_RETRY;
2252
2253 /*
2254 * At task move, charge accounts can be doubly counted. So, it's
2255 * better to wait until the end of task_move if something is going on.
2256 */
2257 if (mem_cgroup_wait_acct_move(mem_over_limit))
2258 return CHARGE_RETRY;
2259
2260 /* If we don't need to call oom-killer at el, return immediately */
2261 if (!oom_check)
2262 return CHARGE_NOMEM;
2263 /* check OOM */
2264 if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask))
2265 return CHARGE_OOM_DIE;
2266
2267 return CHARGE_RETRY;
2268}
2269
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002270/*
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002271 * Unlike exported interface, "oom" parameter is added. if oom==true,
2272 * oom-killer can be invoked.
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002273 */
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002274static int __mem_cgroup_try_charge(struct mm_struct *mm,
Andrea Arcangeliec168512011-01-13 15:46:56 -08002275 gfp_t gfp_mask,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002276 unsigned int nr_pages,
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002277 struct mem_cgroup **ptr,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002278 bool oom)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002279{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002280 unsigned int batch = max(CHARGE_BATCH, nr_pages);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002281 int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002282 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002283 int ret;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002284
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002285 /*
2286 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2287 * in system level. So, allow to go ahead dying process in addition to
2288 * MEMDIE process.
2289 */
2290 if (unlikely(test_thread_flag(TIF_MEMDIE)
2291 || fatal_signal_pending(current)))
2292 goto bypass;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002293
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002294 /*
Hugh Dickins3be91272008-02-07 00:14:19 -08002295 * We always charge the cgroup the mm_struct belongs to.
2296 * The mm_struct's mem_cgroup changes on task migration if the
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002297 * thread group leader migrates. It's possible that mm is not
2298 * set, if so charge the init_mm (happens for pagecache usage).
2299 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002300 if (!*ptr && !mm)
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002301 goto bypass;
2302again:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002303 if (*ptr) { /* css should be a valid one */
2304 memcg = *ptr;
2305 VM_BUG_ON(css_is_removed(&memcg->css));
2306 if (mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002307 goto done;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002308 if (nr_pages == 1 && consume_stock(memcg))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002309 goto done;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002310 css_get(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002311 } else {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002312 struct task_struct *p;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002313
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002314 rcu_read_lock();
2315 p = rcu_dereference(mm->owner);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002316 /*
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002317 * Because we don't have task_lock(), "p" can exit.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002318 * In that case, "memcg" can point to root or p can be NULL with
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002319 * race with swapoff. Then, we have small risk of mis-accouning.
2320 * But such kind of mis-account by race always happens because
2321 * we don't have cgroup_mutex(). It's overkill and we allo that
2322 * small race, here.
2323 * (*) swapoff at el will charge against mm-struct not against
2324 * task-struct. So, mm->owner can be NULL.
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002325 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002326 memcg = mem_cgroup_from_task(p);
2327 if (!memcg || mem_cgroup_is_root(memcg)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002328 rcu_read_unlock();
2329 goto done;
2330 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002331 if (nr_pages == 1 && consume_stock(memcg)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002332 /*
2333 * It seems dagerous to access memcg without css_get().
2334 * But considering how consume_stok works, it's not
2335 * necessary. If consume_stock success, some charges
2336 * from this memcg are cached on this cpu. So, we
2337 * don't need to call css_get()/css_tryget() before
2338 * calling consume_stock().
2339 */
2340 rcu_read_unlock();
2341 goto done;
2342 }
2343 /* after here, we may be blocked. we need to get refcnt */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002344 if (!css_tryget(&memcg->css)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002345 rcu_read_unlock();
2346 goto again;
2347 }
2348 rcu_read_unlock();
2349 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002350
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002351 do {
2352 bool oom_check;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002353
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002354 /* If killed, bypass charge */
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002355 if (fatal_signal_pending(current)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002356 css_put(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002357 goto bypass;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002358 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002359
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002360 oom_check = false;
2361 if (oom && !nr_oom_retries) {
2362 oom_check = true;
2363 nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
2364 }
Balbir Singh6d61ef42009-01-07 18:08:06 -08002365
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002366 ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, oom_check);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002367 switch (ret) {
2368 case CHARGE_OK:
2369 break;
2370 case CHARGE_RETRY: /* not in OOM situation but retry */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002371 batch = nr_pages;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002372 css_put(&memcg->css);
2373 memcg = NULL;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002374 goto again;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002375 case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002376 css_put(&memcg->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002377 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002378 case CHARGE_NOMEM: /* OOM routine works */
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002379 if (!oom) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002380 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002381 goto nomem;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002382 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002383 /* If oom, we never return -ENOMEM */
2384 nr_oom_retries--;
2385 break;
2386 case CHARGE_OOM_DIE: /* Killed by OOM Killer */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002387 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002388 goto bypass;
Balbir Singh66e17072008-02-07 00:13:56 -08002389 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002390 } while (ret != CHARGE_OK);
2391
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002392 if (batch > nr_pages)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002393 refill_stock(memcg, batch - nr_pages);
2394 css_put(&memcg->css);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002395done:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002396 *ptr = memcg;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002397 return 0;
2398nomem:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002399 *ptr = NULL;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002400 return -ENOMEM;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002401bypass:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002402 *ptr = NULL;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002403 return 0;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002404}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002405
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002406/*
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002407 * Somemtimes we have to undo a charge we got by try_charge().
2408 * This function is for that and do uncharge, put css's refcnt.
2409 * gotten by try_charge().
2410 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002411static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg,
Johannes Weinere7018b82011-03-23 16:42:33 -07002412 unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002413{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002414 if (!mem_cgroup_is_root(memcg)) {
Johannes Weinere7018b82011-03-23 16:42:33 -07002415 unsigned long bytes = nr_pages * PAGE_SIZE;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002416
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002417 res_counter_uncharge(&memcg->res, bytes);
Johannes Weinere7018b82011-03-23 16:42:33 -07002418 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002419 res_counter_uncharge(&memcg->memsw, bytes);
Johannes Weinere7018b82011-03-23 16:42:33 -07002420 }
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002421}
2422
2423/*
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002424 * A helper function to get mem_cgroup from ID. must be called under
2425 * rcu_read_lock(). The caller must check css_is_removed() or some if
2426 * it's concern. (dropping refcnt from swap can be called against removed
2427 * memcg.)
2428 */
2429static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
2430{
2431 struct cgroup_subsys_state *css;
2432
2433 /* ID 0 is unused ID */
2434 if (!id)
2435 return NULL;
2436 css = css_lookup(&mem_cgroup_subsys, id);
2437 if (!css)
2438 return NULL;
2439 return container_of(css, struct mem_cgroup, css);
2440}
2441
Wu Fengguange42d9d52009-12-16 12:19:59 +01002442struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002443{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002444 struct mem_cgroup *memcg = NULL;
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002445 struct page_cgroup *pc;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002446 unsigned short id;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002447 swp_entry_t ent;
2448
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002449 VM_BUG_ON(!PageLocked(page));
2450
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002451 pc = lookup_page_cgroup(page);
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002452 lock_page_cgroup(pc);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002453 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002454 memcg = pc->mem_cgroup;
2455 if (memcg && !css_tryget(&memcg->css))
2456 memcg = NULL;
Wu Fengguange42d9d52009-12-16 12:19:59 +01002457 } else if (PageSwapCache(page)) {
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002458 ent.val = page_private(page);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002459 id = lookup_swap_cgroup(ent);
2460 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002461 memcg = mem_cgroup_lookup(id);
2462 if (memcg && !css_tryget(&memcg->css))
2463 memcg = NULL;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002464 rcu_read_unlock();
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002465 }
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002466 unlock_page_cgroup(pc);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002467 return memcg;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002468}
2469
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002470static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg,
Johannes Weiner5564e882011-03-23 16:42:29 -07002471 struct page *page,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002472 unsigned int nr_pages,
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002473 struct page_cgroup *pc,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002474 enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002475{
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002476 lock_page_cgroup(pc);
2477 if (unlikely(PageCgroupUsed(pc))) {
2478 unlock_page_cgroup(pc);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002479 __mem_cgroup_cancel_charge(memcg, nr_pages);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002480 return;
2481 }
2482 /*
2483 * we don't need page_cgroup_lock about tail pages, becase they are not
2484 * accessed by any other context at this point.
2485 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002486 pc->mem_cgroup = memcg;
KAMEZAWA Hiroyuki261fb612009-09-23 15:56:33 -07002487 /*
2488 * We access a page_cgroup asynchronously without lock_page_cgroup().
2489 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2490 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2491 * before USED bit, we need memory barrier here.
2492 * See mem_cgroup_add_lru_list(), etc.
2493 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002494 smp_wmb();
Balbir Singh4b3bde42009-09-23 15:56:32 -07002495 switch (ctype) {
2496 case MEM_CGROUP_CHARGE_TYPE_CACHE:
2497 case MEM_CGROUP_CHARGE_TYPE_SHMEM:
2498 SetPageCgroupCache(pc);
2499 SetPageCgroupUsed(pc);
2500 break;
2501 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
2502 ClearPageCgroupCache(pc);
2503 SetPageCgroupUsed(pc);
2504 break;
2505 default:
2506 break;
2507 }
Hugh Dickins3be91272008-02-07 00:14:19 -08002508
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002509 mem_cgroup_charge_statistics(memcg, PageCgroupCache(pc), nr_pages);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002510 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08002511 /*
2512 * "charge_statistics" updated event counter. Then, check it.
2513 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2514 * if they exceeds softlimit.
2515 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002516 memcg_check_events(memcg, page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002517}
2518
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002519#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2520
2521#define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MOVE_LOCK) |\
2522 (1 << PCG_ACCT_LRU) | (1 << PCG_MIGRATION))
2523/*
2524 * Because tail pages are not marked as "used", set it. We're under
2525 * zone->lru_lock, 'splitting on pmd' and compund_lock.
2526 */
2527void mem_cgroup_split_huge_fixup(struct page *head, struct page *tail)
2528{
2529 struct page_cgroup *head_pc = lookup_page_cgroup(head);
2530 struct page_cgroup *tail_pc = lookup_page_cgroup(tail);
2531 unsigned long flags;
2532
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08002533 if (mem_cgroup_disabled())
2534 return;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002535 /*
KAMEZAWA Hiroyukiece35ca2011-01-20 14:44:24 -08002536 * We have no races with charge/uncharge but will have races with
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002537 * page state accounting.
2538 */
2539 move_lock_page_cgroup(head_pc, &flags);
2540
2541 tail_pc->mem_cgroup = head_pc->mem_cgroup;
2542 smp_wmb(); /* see __commit_charge() */
KAMEZAWA Hiroyukiece35ca2011-01-20 14:44:24 -08002543 if (PageCgroupAcctLRU(head_pc)) {
2544 enum lru_list lru;
2545 struct mem_cgroup_per_zone *mz;
2546
2547 /*
2548 * LRU flags cannot be copied because we need to add tail
2549 *.page to LRU by generic call and our hook will be called.
2550 * We hold lru_lock, then, reduce counter directly.
2551 */
2552 lru = page_lru(head);
Johannes Weiner97a6c372011-03-23 16:42:27 -07002553 mz = page_cgroup_zoneinfo(head_pc->mem_cgroup, head);
KAMEZAWA Hiroyukiece35ca2011-01-20 14:44:24 -08002554 MEM_CGROUP_ZSTAT(mz, lru) -= 1;
2555 }
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002556 tail_pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT;
2557 move_unlock_page_cgroup(head_pc, &flags);
2558}
2559#endif
2560
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002561/**
Johannes Weinerde3638d2011-03-23 16:42:28 -07002562 * mem_cgroup_move_account - move account of the page
Johannes Weiner5564e882011-03-23 16:42:29 -07002563 * @page: the page
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002564 * @nr_pages: number of regular pages (>1 for huge pages)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002565 * @pc: page_cgroup of the page.
2566 * @from: mem_cgroup which the page is moved from.
2567 * @to: mem_cgroup which the page is moved to. @from != @to.
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002568 * @uncharge: whether we should call uncharge and css_put against @from.
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002569 *
2570 * The caller must confirm following.
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002571 * - page is not on LRU (isolate_page() is useful.)
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002572 * - compound_lock is held when nr_pages > 1
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002573 *
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002574 * This function doesn't do "charge" nor css_get to new cgroup. It should be
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002575 * done by a caller(__mem_cgroup_try_charge would be useful). If @uncharge is
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002576 * true, this function does "uncharge" from old cgroup, but it doesn't if
2577 * @uncharge is false, so a caller should do "uncharge".
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002578 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002579static int mem_cgroup_move_account(struct page *page,
2580 unsigned int nr_pages,
2581 struct page_cgroup *pc,
2582 struct mem_cgroup *from,
2583 struct mem_cgroup *to,
2584 bool uncharge)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002585{
Johannes Weinerde3638d2011-03-23 16:42:28 -07002586 unsigned long flags;
2587 int ret;
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002588
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002589 VM_BUG_ON(from == to);
Johannes Weiner5564e882011-03-23 16:42:29 -07002590 VM_BUG_ON(PageLRU(page));
Johannes Weinerde3638d2011-03-23 16:42:28 -07002591 /*
2592 * The page is isolated from LRU. So, collapse function
2593 * will not handle this page. But page splitting can happen.
2594 * Do this check under compound_page_lock(). The caller should
2595 * hold it.
2596 */
2597 ret = -EBUSY;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002598 if (nr_pages > 1 && !PageTransHuge(page))
Johannes Weinerde3638d2011-03-23 16:42:28 -07002599 goto out;
2600
2601 lock_page_cgroup(pc);
2602
2603 ret = -EINVAL;
2604 if (!PageCgroupUsed(pc) || pc->mem_cgroup != from)
2605 goto unlock;
2606
2607 move_lock_page_cgroup(pc, &flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002608
KAMEZAWA Hiroyuki8725d542010-04-06 14:35:05 -07002609 if (PageCgroupFileMapped(pc)) {
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08002610 /* Update mapped_file data for mem_cgroup */
2611 preempt_disable();
2612 __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2613 __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2614 preempt_enable();
Balbir Singhd69b0422009-06-17 16:26:34 -07002615 }
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002616 mem_cgroup_charge_statistics(from, PageCgroupCache(pc), -nr_pages);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002617 if (uncharge)
2618 /* This is not "cancel", but cancel_charge does all we need. */
Johannes Weinere7018b82011-03-23 16:42:33 -07002619 __mem_cgroup_cancel_charge(from, nr_pages);
Balbir Singhd69b0422009-06-17 16:26:34 -07002620
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002621 /* caller should have done css_get */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002622 pc->mem_cgroup = to;
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002623 mem_cgroup_charge_statistics(to, PageCgroupCache(pc), nr_pages);
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002624 /*
2625 * We charges against "to" which may not have any tasks. Then, "to"
2626 * can be under rmdir(). But in current implementation, caller of
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08002627 * this function is just force_empty() and move charge, so it's
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002628 * guaranteed that "to" is never removed. So, we don't check rmdir
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08002629 * status here.
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002630 */
Johannes Weinerde3638d2011-03-23 16:42:28 -07002631 move_unlock_page_cgroup(pc, &flags);
2632 ret = 0;
2633unlock:
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08002634 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08002635 /*
2636 * check events
2637 */
Johannes Weiner5564e882011-03-23 16:42:29 -07002638 memcg_check_events(to, page);
2639 memcg_check_events(from, page);
Johannes Weinerde3638d2011-03-23 16:42:28 -07002640out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002641 return ret;
2642}
2643
2644/*
2645 * move charges to its parent.
2646 */
2647
Johannes Weiner5564e882011-03-23 16:42:29 -07002648static int mem_cgroup_move_parent(struct page *page,
2649 struct page_cgroup *pc,
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002650 struct mem_cgroup *child,
2651 gfp_t gfp_mask)
2652{
2653 struct cgroup *cg = child->css.cgroup;
2654 struct cgroup *pcg = cg->parent;
2655 struct mem_cgroup *parent;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002656 unsigned int nr_pages;
Andrew Morton4be44892011-03-23 16:42:39 -07002657 unsigned long uninitialized_var(flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002658 int ret;
2659
2660 /* Is ROOT ? */
2661 if (!pcg)
2662 return -EINVAL;
2663
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08002664 ret = -EBUSY;
2665 if (!get_page_unless_zero(page))
2666 goto out;
2667 if (isolate_lru_page(page))
2668 goto put;
KAMEZAWA Hiroyuki52dbb902011-01-25 15:07:29 -08002669
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002670 nr_pages = hpage_nr_pages(page);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002671
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002672 parent = mem_cgroup_from_cont(pcg);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002673 ret = __mem_cgroup_try_charge(NULL, gfp_mask, nr_pages, &parent, false);
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002674 if (ret || !parent)
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08002675 goto put_back;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002676
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002677 if (nr_pages > 1)
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002678 flags = compound_lock_irqsave(page);
2679
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002680 ret = mem_cgroup_move_account(page, nr_pages, pc, child, parent, true);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002681 if (ret)
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002682 __mem_cgroup_cancel_charge(parent, nr_pages);
Jesper Juhl8dba4742011-01-25 15:07:24 -08002683
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002684 if (nr_pages > 1)
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002685 compound_unlock_irqrestore(page, flags);
Jesper Juhl8dba4742011-01-25 15:07:24 -08002686put_back:
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002687 putback_lru_page(page);
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08002688put:
Daisuke Nishimura40d58132009-01-15 13:51:12 -08002689 put_page(page);
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08002690out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002691 return ret;
2692}
2693
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002694/*
2695 * Charge the memory controller for page usage.
2696 * Return
2697 * 0 if the charge was successful
2698 * < 0 if the cgroup is over its limit
2699 */
2700static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
Daisuke Nishimura73045c42010-08-10 18:02:59 -07002701 gfp_t gfp_mask, enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002702{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002703 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002704 unsigned int nr_pages = 1;
Johannes Weiner8493ae42011-02-01 15:52:44 -08002705 struct page_cgroup *pc;
2706 bool oom = true;
2707 int ret;
Andrea Arcangeliec168512011-01-13 15:46:56 -08002708
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002709 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002710 nr_pages <<= compound_order(page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002711 VM_BUG_ON(!PageTransHuge(page));
Johannes Weiner8493ae42011-02-01 15:52:44 -08002712 /*
2713 * Never OOM-kill a process for a huge page. The
2714 * fault handler will fall back to regular pages.
2715 */
2716 oom = false;
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002717 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002718
2719 pc = lookup_page_cgroup(page);
Johannes Weineraf4a6622011-03-23 16:42:24 -07002720 BUG_ON(!pc); /* XXX: remove this and move pc lookup into commit */
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002721
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002722 ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom);
2723 if (ret || !memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002724 return ret;
2725
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002726 __mem_cgroup_commit_charge(memcg, page, nr_pages, pc, ctype);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002727 return 0;
2728}
2729
2730int mem_cgroup_newpage_charge(struct page *page,
2731 struct mm_struct *mm, gfp_t gfp_mask)
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08002732{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002733 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07002734 return 0;
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002735 /*
2736 * If already mapped, we don't have to account.
2737 * If page cache, page->mapping has address_space.
2738 * But page->mapping may have out-of-use anon_vma pointer,
2739 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
2740 * is NULL.
2741 */
2742 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
2743 return 0;
2744 if (unlikely(!mm))
2745 mm = &init_mm;
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08002746 return mem_cgroup_charge_common(page, mm, gfp_mask,
Daisuke Nishimura73045c42010-08-10 18:02:59 -07002747 MEM_CGROUP_CHARGE_TYPE_MAPPED);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08002748}
2749
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002750static void
2751__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
2752 enum charge_type ctype);
2753
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002754static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002755__mem_cgroup_commit_charge_lrucare(struct page *page, struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002756 enum charge_type ctype)
2757{
2758 struct page_cgroup *pc = lookup_page_cgroup(page);
2759 /*
2760 * In some case, SwapCache, FUSE(splice_buf->radixtree), the page
2761 * is already on LRU. It means the page may on some other page_cgroup's
2762 * LRU. Take care of it.
2763 */
2764 mem_cgroup_lru_del_before_commit(page);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002765 __mem_cgroup_commit_charge(memcg, page, 1, pc, ctype);
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002766 mem_cgroup_lru_add_after_commit(page);
2767 return;
2768}
2769
Balbir Singhe1a1cd52008-02-07 00:14:02 -08002770int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
2771 gfp_t gfp_mask)
Balbir Singh8697d332008-02-07 00:13:59 -08002772{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002773 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002774 int ret;
2775
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002776 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07002777 return 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002778 if (PageCompound(page))
2779 return 0;
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07002780
Daisuke Nishimura73045c42010-08-10 18:02:59 -07002781 if (unlikely(!mm))
Balbir Singh8697d332008-02-07 00:13:59 -08002782 mm = &init_mm;
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07002783
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002784 if (page_is_file_cache(page)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002785 ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, &memcg, true);
2786 if (ret || !memcg)
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002787 return ret;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002788
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002789 /*
2790 * FUSE reuses pages without going through the final
2791 * put that would remove them from the LRU list, make
2792 * sure that they get relinked properly.
2793 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002794 __mem_cgroup_commit_charge_lrucare(page, memcg,
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002795 MEM_CGROUP_CHARGE_TYPE_CACHE);
2796 return ret;
2797 }
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002798 /* shmem */
2799 if (PageSwapCache(page)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002800 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &memcg);
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002801 if (!ret)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002802 __mem_cgroup_commit_charge_swapin(page, memcg,
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002803 MEM_CGROUP_CHARGE_TYPE_SHMEM);
2804 } else
2805 ret = mem_cgroup_charge_common(page, mm, gfp_mask,
Daisuke Nishimura73045c42010-08-10 18:02:59 -07002806 MEM_CGROUP_CHARGE_TYPE_SHMEM);
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002807
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002808 return ret;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07002809}
2810
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002811/*
2812 * While swap-in, try_charge -> commit or cancel, the page is locked.
2813 * And when try_charge() successfully returns, one refcnt to memcg without
Uwe Kleine-König21ae2952009-10-07 15:21:09 +02002814 * struct page_cgroup is acquired. This refcnt will be consumed by
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002815 * "commit()" or removed by "cancel()"
2816 */
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002817int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
2818 struct page *page,
2819 gfp_t mask, struct mem_cgroup **ptr)
2820{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002821 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002822 int ret;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002823
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -07002824 *ptr = NULL;
2825
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002826 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002827 return 0;
2828
2829 if (!do_swap_account)
2830 goto charge_cur_mm;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002831 /*
2832 * A racing thread's fault, or swapoff, may have already updated
Hugh Dickins407f9c82009-12-14 17:59:30 -08002833 * the pte, and even removed page from swap cache: in those cases
2834 * do_swap_page()'s pte_same() test will fail; but there's also a
2835 * KSM case which does need to charge the page.
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002836 */
2837 if (!PageSwapCache(page))
Hugh Dickins407f9c82009-12-14 17:59:30 -08002838 goto charge_cur_mm;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002839 memcg = try_get_mem_cgroup_from_page(page);
2840 if (!memcg)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002841 goto charge_cur_mm;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002842 *ptr = memcg;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002843 ret = __mem_cgroup_try_charge(NULL, mask, 1, ptr, true);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002844 css_put(&memcg->css);
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002845 return ret;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002846charge_cur_mm:
2847 if (unlikely(!mm))
2848 mm = &init_mm;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002849 return __mem_cgroup_try_charge(mm, mask, 1, ptr, true);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002850}
2851
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002852static void
2853__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
2854 enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002855{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002856 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002857 return;
2858 if (!ptr)
2859 return;
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002860 cgroup_exclude_rmdir(&ptr->css);
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002861
2862 __mem_cgroup_commit_charge_lrucare(page, ptr, ctype);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002863 /*
2864 * Now swap is on-memory. This means this page may be
2865 * counted both as mem and swap....double count.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002866 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2867 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2868 * may call delete_from_swap_cache() before reach here.
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002869 */
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002870 if (do_swap_account && PageSwapCache(page)) {
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002871 swp_entry_t ent = {.val = page_private(page)};
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002872 unsigned short id;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002873 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002874
2875 id = swap_cgroup_record(ent, 0);
2876 rcu_read_lock();
2877 memcg = mem_cgroup_lookup(id);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002878 if (memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002879 /*
2880 * This recorded memcg can be obsolete one. So, avoid
2881 * calling css_tryget
2882 */
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002883 if (!mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -07002884 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002885 mem_cgroup_swap_statistics(memcg, false);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002886 mem_cgroup_put(memcg);
2887 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002888 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002889 }
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002890 /*
2891 * At swapin, we may charge account against cgroup which has no tasks.
2892 * So, rmdir()->pre_destroy() can be called while we do this charge.
2893 * In that case, we need to call pre_destroy() again. check it here.
2894 */
2895 cgroup_release_and_wakeup_rmdir(&ptr->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002896}
2897
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002898void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
2899{
2900 __mem_cgroup_commit_charge_swapin(page, ptr,
2901 MEM_CGROUP_CHARGE_TYPE_MAPPED);
2902}
2903
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002904void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002905{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002906 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002907 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002908 if (!memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002909 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002910 __mem_cgroup_cancel_charge(memcg, 1);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002911}
2912
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002913static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002914 unsigned int nr_pages,
2915 const enum charge_type ctype)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002916{
2917 struct memcg_batch_info *batch = NULL;
2918 bool uncharge_memsw = true;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002919
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002920 /* If swapout, usage of swap doesn't decrease */
2921 if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2922 uncharge_memsw = false;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002923
2924 batch = &current->memcg_batch;
2925 /*
2926 * In usual, we do css_get() when we remember memcg pointer.
2927 * But in this case, we keep res->usage until end of a series of
2928 * uncharges. Then, it's ok to ignore memcg's refcnt.
2929 */
2930 if (!batch->memcg)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002931 batch->memcg = memcg;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002932 /*
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002933 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002934 * In those cases, all pages freed continuously can be expected to be in
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002935 * the same cgroup and we have chance to coalesce uncharges.
2936 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2937 * because we want to do uncharge as soon as possible.
2938 */
2939
2940 if (!batch->do_batch || test_thread_flag(TIF_MEMDIE))
2941 goto direct_uncharge;
2942
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002943 if (nr_pages > 1)
Andrea Arcangeliec168512011-01-13 15:46:56 -08002944 goto direct_uncharge;
2945
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002946 /*
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002947 * In typical case, batch->memcg == mem. This means we can
2948 * merge a series of uncharges to an uncharge of res_counter.
2949 * If not, we uncharge res_counter ony by one.
2950 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002951 if (batch->memcg != memcg)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002952 goto direct_uncharge;
2953 /* remember freed charge and uncharge it later */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07002954 batch->nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002955 if (uncharge_memsw)
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07002956 batch->memsw_nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002957 return;
2958direct_uncharge:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002959 res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002960 if (uncharge_memsw)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002961 res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE);
2962 if (unlikely(batch->memcg != memcg))
2963 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002964 return;
2965}
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002966
Balbir Singh8697d332008-02-07 00:13:59 -08002967/*
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002968 * uncharge if !page_mapped(page)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002969 */
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002970static struct mem_cgroup *
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002971__mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002972{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002973 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002974 unsigned int nr_pages = 1;
2975 struct page_cgroup *pc;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002976
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002977 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002978 return NULL;
Balbir Singh40779602008-04-04 14:29:59 -07002979
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002980 if (PageSwapCache(page))
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002981 return NULL;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002982
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002983 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002984 nr_pages <<= compound_order(page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002985 VM_BUG_ON(!PageTransHuge(page));
2986 }
Balbir Singh8697d332008-02-07 00:13:59 -08002987 /*
Balbir Singh3c541e12008-02-07 00:14:41 -08002988 * Check if our page_cgroup is valid
Balbir Singh8697d332008-02-07 00:13:59 -08002989 */
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002990 pc = lookup_page_cgroup(page);
2991 if (unlikely(!pc || !PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002992 return NULL;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002993
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002994 lock_page_cgroup(pc);
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002995
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002996 memcg = pc->mem_cgroup;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002997
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002998 if (!PageCgroupUsed(pc))
2999 goto unlock_out;
3000
3001 switch (ctype) {
3002 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07003003 case MEM_CGROUP_CHARGE_TYPE_DROP:
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003004 /* See mem_cgroup_prepare_migration() */
3005 if (page_mapped(page) || PageCgroupMigration(pc))
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003006 goto unlock_out;
3007 break;
3008 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
3009 if (!PageAnon(page)) { /* Shared memory */
3010 if (page->mapping && !page_is_file_cache(page))
3011 goto unlock_out;
3012 } else if (page_mapped(page)) /* Anon */
3013 goto unlock_out;
3014 break;
3015 default:
3016 break;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003017 }
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003018
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003019 mem_cgroup_charge_statistics(memcg, PageCgroupCache(pc), -nr_pages);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07003020
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003021 ClearPageCgroupUsed(pc);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08003022 /*
3023 * pc->mem_cgroup is not cleared here. It will be accessed when it's
3024 * freed from LRU. This is safe because uncharged page is expected not
3025 * to be reused (freed soon). Exception is SwapCache, it's handled by
3026 * special functions.
3027 */
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08003028
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003029 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003030 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003031 * even after unlock, we have memcg->res.usage here and this memcg
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003032 * will never be freed.
3033 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003034 memcg_check_events(memcg, page);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003035 if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003036 mem_cgroup_swap_statistics(memcg, true);
3037 mem_cgroup_get(memcg);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003038 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003039 if (!mem_cgroup_is_root(memcg))
3040 mem_cgroup_do_uncharge(memcg, nr_pages, ctype);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003041
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003042 return memcg;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003043
3044unlock_out:
3045 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003046 return NULL;
Balbir Singh3c541e12008-02-07 00:14:41 -08003047}
3048
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003049void mem_cgroup_uncharge_page(struct page *page)
3050{
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003051 /* early check. */
3052 if (page_mapped(page))
3053 return;
3054 if (page->mapping && !PageAnon(page))
3055 return;
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003056 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
3057}
3058
3059void mem_cgroup_uncharge_cache_page(struct page *page)
3060{
3061 VM_BUG_ON(page_mapped(page));
KAMEZAWA Hiroyukib7abea92008-10-18 20:28:09 -07003062 VM_BUG_ON(page->mapping);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003063 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
3064}
3065
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003066/*
3067 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
3068 * In that cases, pages are freed continuously and we can expect pages
3069 * are in the same memcg. All these calls itself limits the number of
3070 * pages freed at once, then uncharge_start/end() is called properly.
3071 * This may be called prural(2) times in a context,
3072 */
3073
3074void mem_cgroup_uncharge_start(void)
3075{
3076 current->memcg_batch.do_batch++;
3077 /* We can do nest. */
3078 if (current->memcg_batch.do_batch == 1) {
3079 current->memcg_batch.memcg = NULL;
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07003080 current->memcg_batch.nr_pages = 0;
3081 current->memcg_batch.memsw_nr_pages = 0;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003082 }
3083}
3084
3085void mem_cgroup_uncharge_end(void)
3086{
3087 struct memcg_batch_info *batch = &current->memcg_batch;
3088
3089 if (!batch->do_batch)
3090 return;
3091
3092 batch->do_batch--;
3093 if (batch->do_batch) /* If stacked, do nothing. */
3094 return;
3095
3096 if (!batch->memcg)
3097 return;
3098 /*
3099 * This "batch->memcg" is valid without any css_get/put etc...
3100 * bacause we hide charges behind us.
3101 */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07003102 if (batch->nr_pages)
3103 res_counter_uncharge(&batch->memcg->res,
3104 batch->nr_pages * PAGE_SIZE);
3105 if (batch->memsw_nr_pages)
3106 res_counter_uncharge(&batch->memcg->memsw,
3107 batch->memsw_nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003108 memcg_oom_recover(batch->memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003109 /* forget this pointer (for sanity check) */
3110 batch->memcg = NULL;
3111}
3112
Daisuke Nishimurae767e052009-05-28 14:34:28 -07003113#ifdef CONFIG_SWAP
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003114/*
Daisuke Nishimurae767e052009-05-28 14:34:28 -07003115 * called after __delete_from_swap_cache() and drop "page" account.
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003116 * memcg information is recorded to swap_cgroup of "ent"
3117 */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07003118void
3119mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003120{
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003121 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07003122 int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003123
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07003124 if (!swapout) /* this was a swap cache but the swap is unused ! */
3125 ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
3126
3127 memcg = __mem_cgroup_uncharge_common(page, ctype);
3128
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003129 /*
3130 * record memcg information, if swapout && memcg != NULL,
3131 * mem_cgroup_get() was called in uncharge().
3132 */
3133 if (do_swap_account && swapout && memcg)
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003134 swap_cgroup_record(ent, css_id(&memcg->css));
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003135}
Daisuke Nishimurae767e052009-05-28 14:34:28 -07003136#endif
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003137
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003138#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3139/*
3140 * called from swap_entry_free(). remove record in swap_cgroup and
3141 * uncharge "memsw" account.
3142 */
3143void mem_cgroup_uncharge_swap(swp_entry_t ent)
3144{
3145 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003146 unsigned short id;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003147
3148 if (!do_swap_account)
3149 return;
3150
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003151 id = swap_cgroup_record(ent, 0);
3152 rcu_read_lock();
3153 memcg = mem_cgroup_lookup(id);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003154 if (memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003155 /*
3156 * We uncharge this because swap is freed.
3157 * This memcg can be obsolete one. We avoid calling css_tryget
3158 */
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003159 if (!mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -07003160 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003161 mem_cgroup_swap_statistics(memcg, false);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003162 mem_cgroup_put(memcg);
3163 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003164 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003165}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003166
3167/**
3168 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3169 * @entry: swap entry to be moved
3170 * @from: mem_cgroup which the entry is moved from
3171 * @to: mem_cgroup which the entry is moved to
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003172 * @need_fixup: whether we should fixup res_counters and refcounts.
Daisuke Nishimura02491442010-03-10 15:22:17 -08003173 *
3174 * It succeeds only when the swap_cgroup's record for this entry is the same
3175 * as the mem_cgroup's id of @from.
3176 *
3177 * Returns 0 on success, -EINVAL on failure.
3178 *
3179 * The caller must have charged to @to, IOW, called res_counter_charge() about
3180 * both res and memsw, and called css_get().
3181 */
3182static int mem_cgroup_move_swap_account(swp_entry_t entry,
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003183 struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
Daisuke Nishimura02491442010-03-10 15:22:17 -08003184{
3185 unsigned short old_id, new_id;
3186
3187 old_id = css_id(&from->css);
3188 new_id = css_id(&to->css);
3189
3190 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08003191 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08003192 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003193 /*
3194 * This function is only called from task migration context now.
3195 * It postpones res_counter and refcount handling till the end
3196 * of task migration(mem_cgroup_clear_mc()) for performance
3197 * improvement. But we cannot postpone mem_cgroup_get(to)
3198 * because if the process that has been moved to @to does
3199 * swap-in, the refcount of @to might be decreased to 0.
3200 */
Daisuke Nishimura02491442010-03-10 15:22:17 -08003201 mem_cgroup_get(to);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003202 if (need_fixup) {
3203 if (!mem_cgroup_is_root(from))
3204 res_counter_uncharge(&from->memsw, PAGE_SIZE);
3205 mem_cgroup_put(from);
3206 /*
3207 * we charged both to->res and to->memsw, so we should
3208 * uncharge to->res.
3209 */
3210 if (!mem_cgroup_is_root(to))
3211 res_counter_uncharge(&to->res, PAGE_SIZE);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003212 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08003213 return 0;
3214 }
3215 return -EINVAL;
3216}
3217#else
3218static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003219 struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
Daisuke Nishimura02491442010-03-10 15:22:17 -08003220{
3221 return -EINVAL;
3222}
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003223#endif
3224
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003225/*
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003226 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3227 * page belongs to.
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003228 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003229int mem_cgroup_prepare_migration(struct page *page,
Miklos Szeredief6a3c62011-03-22 16:30:52 -07003230 struct page *newpage, struct mem_cgroup **ptr, gfp_t gfp_mask)
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003231{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003232 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003233 struct page_cgroup *pc;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003234 enum charge_type ctype;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003235 int ret = 0;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08003236
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -07003237 *ptr = NULL;
3238
Andrea Arcangeliec168512011-01-13 15:46:56 -08003239 VM_BUG_ON(PageTransHuge(page));
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08003240 if (mem_cgroup_disabled())
Balbir Singh40779602008-04-04 14:29:59 -07003241 return 0;
3242
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003243 pc = lookup_page_cgroup(page);
3244 lock_page_cgroup(pc);
3245 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003246 memcg = pc->mem_cgroup;
3247 css_get(&memcg->css);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003248 /*
3249 * At migrating an anonymous page, its mapcount goes down
3250 * to 0 and uncharge() will be called. But, even if it's fully
3251 * unmapped, migration may fail and this page has to be
3252 * charged again. We set MIGRATION flag here and delay uncharge
3253 * until end_migration() is called
3254 *
3255 * Corner Case Thinking
3256 * A)
3257 * When the old page was mapped as Anon and it's unmap-and-freed
3258 * while migration was ongoing.
3259 * If unmap finds the old page, uncharge() of it will be delayed
3260 * until end_migration(). If unmap finds a new page, it's
3261 * uncharged when it make mapcount to be 1->0. If unmap code
3262 * finds swap_migration_entry, the new page will not be mapped
3263 * and end_migration() will find it(mapcount==0).
3264 *
3265 * B)
3266 * When the old page was mapped but migraion fails, the kernel
3267 * remaps it. A charge for it is kept by MIGRATION flag even
3268 * if mapcount goes down to 0. We can do remap successfully
3269 * without charging it again.
3270 *
3271 * C)
3272 * The "old" page is under lock_page() until the end of
3273 * migration, so, the old page itself will not be swapped-out.
3274 * If the new page is swapped out before end_migraton, our
3275 * hook to usual swap-out path will catch the event.
3276 */
3277 if (PageAnon(page))
3278 SetPageCgroupMigration(pc);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08003279 }
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003280 unlock_page_cgroup(pc);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003281 /*
3282 * If the page is not charged at this point,
3283 * we return here.
3284 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003285 if (!memcg)
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003286 return 0;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003287
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003288 *ptr = memcg;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003289 ret = __mem_cgroup_try_charge(NULL, gfp_mask, 1, ptr, false);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003290 css_put(&memcg->css);/* drop extra refcnt */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003291 if (ret || *ptr == NULL) {
3292 if (PageAnon(page)) {
3293 lock_page_cgroup(pc);
3294 ClearPageCgroupMigration(pc);
3295 unlock_page_cgroup(pc);
3296 /*
3297 * The old page may be fully unmapped while we kept it.
3298 */
3299 mem_cgroup_uncharge_page(page);
3300 }
3301 return -ENOMEM;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003302 }
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003303 /*
3304 * We charge new page before it's used/mapped. So, even if unlock_page()
3305 * is called before end_migration, we can catch all events on this new
3306 * page. In the case new page is migrated but not remapped, new page's
3307 * mapcount will be finally 0 and we call uncharge in end_migration().
3308 */
3309 pc = lookup_page_cgroup(newpage);
3310 if (PageAnon(page))
3311 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
3312 else if (page_is_file_cache(page))
3313 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
3314 else
3315 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003316 __mem_cgroup_commit_charge(memcg, page, 1, pc, ctype);
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003317 return ret;
3318}
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -08003319
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003320/* remove redundant charge if migration failed*/
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003321void mem_cgroup_end_migration(struct mem_cgroup *memcg,
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08003322 struct page *oldpage, struct page *newpage, bool migration_ok)
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003323{
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003324 struct page *used, *unused;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003325 struct page_cgroup *pc;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003326
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003327 if (!memcg)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003328 return;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003329 /* blocks rmdir() */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003330 cgroup_exclude_rmdir(&memcg->css);
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08003331 if (!migration_ok) {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003332 used = oldpage;
3333 unused = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003334 } else {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003335 used = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003336 unused = oldpage;
3337 }
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003338 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003339 * We disallowed uncharge of pages under migration because mapcount
3340 * of the page goes down to zero, temporarly.
3341 * Clear the flag and check the page should be charged.
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003342 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003343 pc = lookup_page_cgroup(oldpage);
3344 lock_page_cgroup(pc);
3345 ClearPageCgroupMigration(pc);
3346 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003347
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003348 __mem_cgroup_uncharge_common(unused, MEM_CGROUP_CHARGE_TYPE_FORCE);
3349
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003350 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003351 * If a page is a file cache, radix-tree replacement is very atomic
3352 * and we can skip this check. When it was an Anon page, its mapcount
3353 * goes down to 0. But because we added MIGRATION flage, it's not
3354 * uncharged yet. There are several case but page->mapcount check
3355 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3356 * check. (see prepare_charge() also)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003357 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003358 if (PageAnon(used))
3359 mem_cgroup_uncharge_page(used);
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07003360 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003361 * At migration, we may charge account against cgroup which has no
3362 * tasks.
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07003363 * So, rmdir()->pre_destroy() can be called while we do this charge.
3364 * In that case, we need to call pre_destroy() again. check it here.
3365 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003366 cgroup_release_and_wakeup_rmdir(&memcg->css);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003367}
Pavel Emelianov78fb7462008-02-07 00:13:51 -08003368
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003369#ifdef CONFIG_DEBUG_VM
3370static struct page_cgroup *lookup_page_cgroup_used(struct page *page)
3371{
3372 struct page_cgroup *pc;
3373
3374 pc = lookup_page_cgroup(page);
3375 if (likely(pc) && PageCgroupUsed(pc))
3376 return pc;
3377 return NULL;
3378}
3379
3380bool mem_cgroup_bad_page_check(struct page *page)
3381{
3382 if (mem_cgroup_disabled())
3383 return false;
3384
3385 return lookup_page_cgroup_used(page) != NULL;
3386}
3387
3388void mem_cgroup_print_bad_page(struct page *page)
3389{
3390 struct page_cgroup *pc;
3391
3392 pc = lookup_page_cgroup_used(page);
3393 if (pc) {
3394 int ret = -1;
3395 char *path;
3396
3397 printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p",
3398 pc, pc->flags, pc->mem_cgroup);
3399
3400 path = kmalloc(PATH_MAX, GFP_KERNEL);
3401 if (path) {
3402 rcu_read_lock();
3403 ret = cgroup_path(pc->mem_cgroup->css.cgroup,
3404 path, PATH_MAX);
3405 rcu_read_unlock();
3406 }
3407
3408 printk(KERN_CONT "(%s)\n",
3409 (ret < 0) ? "cannot get the path" : path);
3410 kfree(path);
3411 }
3412}
3413#endif
3414
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003415static DEFINE_MUTEX(set_limit_mutex);
3416
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08003417static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003418 unsigned long long val)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003419{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003420 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003421 u64 memswlimit, memlimit;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003422 int ret = 0;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003423 int children = mem_cgroup_count_children(memcg);
3424 u64 curusage, oldusage;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003425 int enlarge;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003426
3427 /*
3428 * For keeping hierarchical_reclaim simple, how long we should retry
3429 * is depends on callers. We set our retry-count to be function
3430 * of # of children which we should visit in this loop.
3431 */
3432 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
3433
3434 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003435
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003436 enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003437 while (retry_count) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003438 if (signal_pending(current)) {
3439 ret = -EINTR;
3440 break;
3441 }
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003442 /*
3443 * Rather than hide all in some function, I do this in
3444 * open coded manner. You see what this really does.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003445 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003446 */
3447 mutex_lock(&set_limit_mutex);
3448 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3449 if (memswlimit < val) {
3450 ret = -EINVAL;
3451 mutex_unlock(&set_limit_mutex);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003452 break;
3453 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003454
3455 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3456 if (memlimit < val)
3457 enlarge = 1;
3458
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003459 ret = res_counter_set_limit(&memcg->res, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07003460 if (!ret) {
3461 if (memswlimit == val)
3462 memcg->memsw_is_minimum = true;
3463 else
3464 memcg->memsw_is_minimum = false;
3465 }
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003466 mutex_unlock(&set_limit_mutex);
3467
3468 if (!ret)
3469 break;
3470
Bob Liuaa20d482009-12-15 16:47:14 -08003471 mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL,
Ying Han0ae5e892011-05-26 16:25:25 -07003472 MEM_CGROUP_RECLAIM_SHRINK,
3473 NULL);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003474 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
3475 /* Usage is reduced ? */
3476 if (curusage >= oldusage)
3477 retry_count--;
3478 else
3479 oldusage = curusage;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003480 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003481 if (!ret && enlarge)
3482 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08003483
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003484 return ret;
3485}
3486
Li Zefan338c8432009-06-17 16:27:15 -07003487static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
3488 unsigned long long val)
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003489{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003490 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003491 u64 memlimit, memswlimit, oldusage, curusage;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003492 int children = mem_cgroup_count_children(memcg);
3493 int ret = -EBUSY;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003494 int enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003495
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003496 /* see mem_cgroup_resize_res_limit */
3497 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
3498 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003499 while (retry_count) {
3500 if (signal_pending(current)) {
3501 ret = -EINTR;
3502 break;
3503 }
3504 /*
3505 * Rather than hide all in some function, I do this in
3506 * open coded manner. You see what this really does.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003507 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003508 */
3509 mutex_lock(&set_limit_mutex);
3510 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3511 if (memlimit > val) {
3512 ret = -EINVAL;
3513 mutex_unlock(&set_limit_mutex);
3514 break;
3515 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003516 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3517 if (memswlimit < val)
3518 enlarge = 1;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003519 ret = res_counter_set_limit(&memcg->memsw, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07003520 if (!ret) {
3521 if (memlimit == val)
3522 memcg->memsw_is_minimum = true;
3523 else
3524 memcg->memsw_is_minimum = false;
3525 }
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003526 mutex_unlock(&set_limit_mutex);
3527
3528 if (!ret)
3529 break;
3530
Balbir Singh4e416952009-09-23 15:56:39 -07003531 mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL,
Balbir Singh75822b42009-09-23 15:56:38 -07003532 MEM_CGROUP_RECLAIM_NOSWAP |
Ying Han0ae5e892011-05-26 16:25:25 -07003533 MEM_CGROUP_RECLAIM_SHRINK,
3534 NULL);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003535 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003536 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003537 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003538 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003539 else
3540 oldusage = curusage;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003541 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003542 if (!ret && enlarge)
3543 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003544 return ret;
3545}
3546
Balbir Singh4e416952009-09-23 15:56:39 -07003547unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
Ying Han0ae5e892011-05-26 16:25:25 -07003548 gfp_t gfp_mask,
3549 unsigned long *total_scanned)
Balbir Singh4e416952009-09-23 15:56:39 -07003550{
3551 unsigned long nr_reclaimed = 0;
3552 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
3553 unsigned long reclaimed;
3554 int loop = 0;
3555 struct mem_cgroup_tree_per_zone *mctz;
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -07003556 unsigned long long excess;
Ying Han0ae5e892011-05-26 16:25:25 -07003557 unsigned long nr_scanned;
Balbir Singh4e416952009-09-23 15:56:39 -07003558
3559 if (order > 0)
3560 return 0;
3561
KOSAKI Motohiro00918b62010-08-10 18:03:05 -07003562 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
Balbir Singh4e416952009-09-23 15:56:39 -07003563 /*
3564 * This loop can run a while, specially if mem_cgroup's continuously
3565 * keep exceeding their soft limit and putting the system under
3566 * pressure
3567 */
3568 do {
3569 if (next_mz)
3570 mz = next_mz;
3571 else
3572 mz = mem_cgroup_largest_soft_limit_node(mctz);
3573 if (!mz)
3574 break;
3575
Ying Han0ae5e892011-05-26 16:25:25 -07003576 nr_scanned = 0;
Balbir Singh4e416952009-09-23 15:56:39 -07003577 reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone,
3578 gfp_mask,
Ying Han0ae5e892011-05-26 16:25:25 -07003579 MEM_CGROUP_RECLAIM_SOFT,
3580 &nr_scanned);
Balbir Singh4e416952009-09-23 15:56:39 -07003581 nr_reclaimed += reclaimed;
Ying Han0ae5e892011-05-26 16:25:25 -07003582 *total_scanned += nr_scanned;
Balbir Singh4e416952009-09-23 15:56:39 -07003583 spin_lock(&mctz->lock);
3584
3585 /*
3586 * If we failed to reclaim anything from this memory cgroup
3587 * it is time to move on to the next cgroup
3588 */
3589 next_mz = NULL;
3590 if (!reclaimed) {
3591 do {
3592 /*
3593 * Loop until we find yet another one.
3594 *
3595 * By the time we get the soft_limit lock
3596 * again, someone might have aded the
3597 * group back on the RB tree. Iterate to
3598 * make sure we get a different mem.
3599 * mem_cgroup_largest_soft_limit_node returns
3600 * NULL if no other cgroup is present on
3601 * the tree
3602 */
3603 next_mz =
3604 __mem_cgroup_largest_soft_limit_node(mctz);
Michal Hocko39cc98f2011-05-26 16:25:28 -07003605 if (next_mz == mz)
Balbir Singh4e416952009-09-23 15:56:39 -07003606 css_put(&next_mz->mem->css);
Michal Hocko39cc98f2011-05-26 16:25:28 -07003607 else /* next_mz == NULL or other memcg */
Balbir Singh4e416952009-09-23 15:56:39 -07003608 break;
3609 } while (1);
3610 }
Balbir Singh4e416952009-09-23 15:56:39 -07003611 __mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -07003612 excess = res_counter_soft_limit_excess(&mz->mem->res);
Balbir Singh4e416952009-09-23 15:56:39 -07003613 /*
3614 * One school of thought says that we should not add
3615 * back the node to the tree if reclaim returns 0.
3616 * But our reclaim could return 0, simply because due
3617 * to priority we are exposing a smaller subset of
3618 * memory to reclaim from. Consider this as a longer
3619 * term TODO.
3620 */
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -07003621 /* If excess == 0, no tree ops */
3622 __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess);
Balbir Singh4e416952009-09-23 15:56:39 -07003623 spin_unlock(&mctz->lock);
3624 css_put(&mz->mem->css);
3625 loop++;
3626 /*
3627 * Could not reclaim anything and there are no more
3628 * mem cgroups to try or we seem to be looping without
3629 * reclaiming anything.
3630 */
3631 if (!nr_reclaimed &&
3632 (next_mz == NULL ||
3633 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
3634 break;
3635 } while (!nr_reclaimed);
3636 if (next_mz)
3637 css_put(&next_mz->mem->css);
3638 return nr_reclaimed;
3639}
3640
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07003641/*
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003642 * This routine traverse page_cgroup in given list and drop them all.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003643 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
3644 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003645static int mem_cgroup_force_empty_list(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003646 int node, int zid, enum lru_list lru)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003647{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003648 struct zone *zone;
3649 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003650 struct page_cgroup *pc, *busy;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003651 unsigned long flags, loop;
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -08003652 struct list_head *list;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003653 int ret = 0;
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -08003654
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003655 zone = &NODE_DATA(node)->node_zones[zid];
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003656 mz = mem_cgroup_zoneinfo(memcg, node, zid);
Christoph Lameterb69408e2008-10-18 20:26:14 -07003657 list = &mz->lists[lru];
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003658
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003659 loop = MEM_CGROUP_ZSTAT(mz, lru);
3660 /* give some margin against EBUSY etc...*/
3661 loop += 256;
3662 busy = NULL;
3663 while (loop--) {
Johannes Weiner5564e882011-03-23 16:42:29 -07003664 struct page *page;
3665
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003666 ret = 0;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003667 spin_lock_irqsave(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003668 if (list_empty(list)) {
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003669 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003670 break;
3671 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003672 pc = list_entry(list->prev, struct page_cgroup, lru);
3673 if (busy == pc) {
3674 list_move(&pc->lru, list);
Thiago Farina648bcc72010-03-05 13:42:04 -08003675 busy = NULL;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003676 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003677 continue;
3678 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003679 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003680
Johannes Weiner6b3ae582011-03-23 16:42:30 -07003681 page = lookup_cgroup_page(pc);
Johannes Weiner5564e882011-03-23 16:42:29 -07003682
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003683 ret = mem_cgroup_move_parent(page, pc, memcg, GFP_KERNEL);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003684 if (ret == -ENOMEM)
3685 break;
3686
3687 if (ret == -EBUSY || ret == -EINVAL) {
3688 /* found lock contention or "pc" is obsolete. */
3689 busy = pc;
3690 cond_resched();
3691 } else
3692 busy = NULL;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003693 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003694
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003695 if (!ret && !list_empty(list))
3696 return -EBUSY;
3697 return ret;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003698}
3699
3700/*
3701 * make mem_cgroup's charge to be 0 if there is no task.
3702 * This enables deleting this mem_cgroup.
3703 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003704static int mem_cgroup_force_empty(struct mem_cgroup *memcg, bool free_all)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003705{
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003706 int ret;
3707 int node, zid, shrink;
3708 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003709 struct cgroup *cgrp = memcg->css.cgroup;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08003710
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003711 css_get(&memcg->css);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003712
3713 shrink = 0;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003714 /* should free all ? */
3715 if (free_all)
3716 goto try_to_free;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003717move_account:
Daisuke Nishimurafce66472010-01-15 17:01:30 -08003718 do {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003719 ret = -EBUSY;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003720 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003721 goto out;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003722 ret = -EINTR;
3723 if (signal_pending(current))
3724 goto out;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003725 /* This is for making all *used* pages to be on LRU. */
3726 lru_add_drain_all();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003727 drain_all_stock_sync(memcg);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003728 ret = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003729 mem_cgroup_start_move(memcg);
KAMEZAWA Hiroyuki299b4ea2009-01-29 14:25:17 -08003730 for_each_node_state(node, N_HIGH_MEMORY) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003731 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
Christoph Lameterb69408e2008-10-18 20:26:14 -07003732 enum lru_list l;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003733 for_each_lru(l) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003734 ret = mem_cgroup_force_empty_list(memcg,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003735 node, zid, l);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003736 if (ret)
3737 break;
3738 }
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08003739 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003740 if (ret)
3741 break;
3742 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003743 mem_cgroup_end_move(memcg);
3744 memcg_oom_recover(memcg);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003745 /* it seems parent cgroup doesn't have enough mem */
3746 if (ret == -ENOMEM)
3747 goto try_to_free;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003748 cond_resched();
Daisuke Nishimurafce66472010-01-15 17:01:30 -08003749 /* "ret" should also be checked to ensure all lists are empty. */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003750 } while (memcg->res.usage > 0 || ret);
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003751out:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003752 css_put(&memcg->css);
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003753 return ret;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003754
3755try_to_free:
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003756 /* returns EBUSY if there is a task or if we come here twice. */
3757 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003758 ret = -EBUSY;
3759 goto out;
3760 }
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003761 /* we call try-to-free pages for make this cgroup empty */
3762 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003763 /* try to free all pages in this cgroup */
3764 shrink = 1;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003765 while (nr_retries && memcg->res.usage > 0) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003766 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003767
3768 if (signal_pending(current)) {
3769 ret = -EINTR;
3770 goto out;
3771 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003772 progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL,
Johannes Weiner185efc02011-09-14 16:21:58 -07003773 false);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003774 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003775 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003776 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02003777 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003778 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003779
3780 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003781 lru_add_drain();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003782 /* try move_account...there may be some *locked* pages. */
Daisuke Nishimurafce66472010-01-15 17:01:30 -08003783 goto move_account;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003784}
3785
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003786int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
3787{
3788 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
3789}
3790
3791
Balbir Singh18f59ea2009-01-07 18:08:07 -08003792static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
3793{
3794 return mem_cgroup_from_cont(cont)->use_hierarchy;
3795}
3796
3797static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
3798 u64 val)
3799{
3800 int retval = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003801 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Balbir Singh18f59ea2009-01-07 18:08:07 -08003802 struct cgroup *parent = cont->parent;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003803 struct mem_cgroup *parent_memcg = NULL;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003804
3805 if (parent)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003806 parent_memcg = mem_cgroup_from_cont(parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08003807
3808 cgroup_lock();
3809 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02003810 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08003811 * in the child subtrees. If it is unset, then the change can
3812 * occur, provided the current cgroup has no children.
3813 *
3814 * For the root cgroup, parent_mem is NULL, we allow value to be
3815 * set if there are no children.
3816 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003817 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08003818 (val == 1 || val == 0)) {
3819 if (list_empty(&cont->children))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003820 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003821 else
3822 retval = -EBUSY;
3823 } else
3824 retval = -EINVAL;
3825 cgroup_unlock();
3826
3827 return retval;
3828}
3829
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003830
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003831static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -07003832 enum mem_cgroup_stat_index idx)
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003833{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003834 struct mem_cgroup *iter;
Johannes Weiner7a159cc2011-03-23 16:42:38 -07003835 long val = 0;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003836
Johannes Weiner7a159cc2011-03-23 16:42:38 -07003837 /* Per-cpu values can be negative, use a signed accumulator */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003838 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003839 val += mem_cgroup_read_stat(iter, idx);
3840
3841 if (val < 0) /* race ? */
3842 val = 0;
3843 return val;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003844}
3845
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003846static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003847{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003848 u64 val;
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003849
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003850 if (!mem_cgroup_is_root(memcg)) {
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003851 if (!swap)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003852 return res_counter_read_u64(&memcg->res, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003853 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003854 return res_counter_read_u64(&memcg->memsw, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003855 }
3856
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003857 val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE);
3858 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003859
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003860 if (swap)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003861 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAPOUT);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003862
3863 return val << PAGE_SHIFT;
3864}
3865
Paul Menage2c3daa72008-04-29 00:59:58 -07003866static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003867{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003868 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003869 u64 val;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003870 int type, name;
3871
3872 type = MEMFILE_TYPE(cft->private);
3873 name = MEMFILE_ATTR(cft->private);
3874 switch (type) {
3875 case _MEM:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003876 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003877 val = mem_cgroup_usage(memcg, false);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003878 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003879 val = res_counter_read_u64(&memcg->res, name);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003880 break;
3881 case _MEMSWAP:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003882 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003883 val = mem_cgroup_usage(memcg, true);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003884 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003885 val = res_counter_read_u64(&memcg->memsw, name);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003886 break;
3887 default:
3888 BUG();
3889 break;
3890 }
3891 return val;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003892}
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003893/*
3894 * The user of this function is...
3895 * RES_LIMIT.
3896 */
Paul Menage856c13a2008-07-25 01:47:04 -07003897static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
3898 const char *buffer)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003899{
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003900 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003901 int type, name;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003902 unsigned long long val;
3903 int ret;
3904
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003905 type = MEMFILE_TYPE(cft->private);
3906 name = MEMFILE_ATTR(cft->private);
3907 switch (name) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003908 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07003909 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
3910 ret = -EINVAL;
3911 break;
3912 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003913 /* This function does all necessary parse...reuse it */
3914 ret = res_counter_memparse_write_strategy(buffer, &val);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003915 if (ret)
3916 break;
3917 if (type == _MEM)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003918 ret = mem_cgroup_resize_limit(memcg, val);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003919 else
3920 ret = mem_cgroup_resize_memsw_limit(memcg, val);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003921 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07003922 case RES_SOFT_LIMIT:
3923 ret = res_counter_memparse_write_strategy(buffer, &val);
3924 if (ret)
3925 break;
3926 /*
3927 * For memsw, soft limits are hard to implement in terms
3928 * of semantics, for now, we support soft limits for
3929 * control without swap
3930 */
3931 if (type == _MEM)
3932 ret = res_counter_set_soft_limit(&memcg->res, val);
3933 else
3934 ret = -EINVAL;
3935 break;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003936 default:
3937 ret = -EINVAL; /* should be BUG() ? */
3938 break;
3939 }
3940 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003941}
3942
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003943static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
3944 unsigned long long *mem_limit, unsigned long long *memsw_limit)
3945{
3946 struct cgroup *cgroup;
3947 unsigned long long min_limit, min_memsw_limit, tmp;
3948
3949 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3950 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3951 cgroup = memcg->css.cgroup;
3952 if (!memcg->use_hierarchy)
3953 goto out;
3954
3955 while (cgroup->parent) {
3956 cgroup = cgroup->parent;
3957 memcg = mem_cgroup_from_cont(cgroup);
3958 if (!memcg->use_hierarchy)
3959 break;
3960 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
3961 min_limit = min(min_limit, tmp);
3962 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3963 min_memsw_limit = min(min_memsw_limit, tmp);
3964 }
3965out:
3966 *mem_limit = min_limit;
3967 *memsw_limit = min_memsw_limit;
3968 return;
3969}
3970
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003971static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003972{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003973 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003974 int type, name;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003975
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003976 memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003977 type = MEMFILE_TYPE(event);
3978 name = MEMFILE_ATTR(event);
3979 switch (name) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003980 case RES_MAX_USAGE:
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003981 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003982 res_counter_reset_max(&memcg->res);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003983 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003984 res_counter_reset_max(&memcg->memsw);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003985 break;
3986 case RES_FAILCNT:
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003987 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003988 res_counter_reset_failcnt(&memcg->res);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003989 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003990 res_counter_reset_failcnt(&memcg->memsw);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003991 break;
3992 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07003993
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -07003994 return 0;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003995}
3996
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003997static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp,
3998 struct cftype *cft)
3999{
4000 return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate;
4001}
4002
Daisuke Nishimura02491442010-03-10 15:22:17 -08004003#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004004static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
4005 struct cftype *cft, u64 val)
4006{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004007 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004008
4009 if (val >= (1 << NR_MOVE_TYPE))
4010 return -EINVAL;
4011 /*
4012 * We check this value several times in both in can_attach() and
4013 * attach(), so we need cgroup lock to prevent this value from being
4014 * inconsistent.
4015 */
4016 cgroup_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004017 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004018 cgroup_unlock();
4019
4020 return 0;
4021}
Daisuke Nishimura02491442010-03-10 15:22:17 -08004022#else
4023static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
4024 struct cftype *cft, u64 val)
4025{
4026 return -ENOSYS;
4027}
4028#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004029
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004030
4031/* For read statistics */
4032enum {
4033 MCS_CACHE,
4034 MCS_RSS,
KAMEZAWA Hiroyukid8046582009-12-15 16:47:09 -08004035 MCS_FILE_MAPPED,
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004036 MCS_PGPGIN,
4037 MCS_PGPGOUT,
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004038 MCS_SWAP,
Ying Han456f9982011-05-26 16:25:38 -07004039 MCS_PGFAULT,
4040 MCS_PGMAJFAULT,
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004041 MCS_INACTIVE_ANON,
4042 MCS_ACTIVE_ANON,
4043 MCS_INACTIVE_FILE,
4044 MCS_ACTIVE_FILE,
4045 MCS_UNEVICTABLE,
4046 NR_MCS_STAT,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004047};
4048
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004049struct mcs_total_stat {
4050 s64 stat[NR_MCS_STAT];
4051};
4052
4053struct {
4054 char *local_name;
4055 char *total_name;
4056} memcg_stat_strings[NR_MCS_STAT] = {
4057 {"cache", "total_cache"},
4058 {"rss", "total_rss"},
Balbir Singhd69b0422009-06-17 16:26:34 -07004059 {"mapped_file", "total_mapped_file"},
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004060 {"pgpgin", "total_pgpgin"},
4061 {"pgpgout", "total_pgpgout"},
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004062 {"swap", "total_swap"},
Ying Han456f9982011-05-26 16:25:38 -07004063 {"pgfault", "total_pgfault"},
4064 {"pgmajfault", "total_pgmajfault"},
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004065 {"inactive_anon", "total_inactive_anon"},
4066 {"active_anon", "total_active_anon"},
4067 {"inactive_file", "total_inactive_file"},
4068 {"active_file", "total_active_file"},
4069 {"unevictable", "total_unevictable"}
4070};
4071
4072
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004073static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004074mem_cgroup_get_local_stat(struct mem_cgroup *memcg, struct mcs_total_stat *s)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004075{
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004076 s64 val;
4077
4078 /* per cpu stat */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004079 val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_CACHE);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004080 s->stat[MCS_CACHE] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004081 val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_RSS);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004082 s->stat[MCS_RSS] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004083 val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED);
KAMEZAWA Hiroyukid8046582009-12-15 16:47:09 -08004084 s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004085 val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGPGIN);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004086 s->stat[MCS_PGPGIN] += val;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004087 val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGPGOUT);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004088 s->stat[MCS_PGPGOUT] += val;
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004089 if (do_swap_account) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004090 val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_SWAPOUT);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004091 s->stat[MCS_SWAP] += val * PAGE_SIZE;
4092 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004093 val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGFAULT);
Ying Han456f9982011-05-26 16:25:38 -07004094 s->stat[MCS_PGFAULT] += val;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004095 val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07004096 s->stat[MCS_PGMAJFAULT] += val;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004097
4098 /* per zone stat */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004099 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_INACTIVE_ANON));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004100 s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004101 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_ACTIVE_ANON));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004102 s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004103 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_INACTIVE_FILE));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004104 s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004105 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_ACTIVE_FILE));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004106 s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004107 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004108 s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004109}
4110
4111static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004112mem_cgroup_get_total_stat(struct mem_cgroup *memcg, struct mcs_total_stat *s)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004113{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004114 struct mem_cgroup *iter;
4115
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004116 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004117 mem_cgroup_get_local_stat(iter, s);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004118}
4119
Ying Han406eb0c2011-05-26 16:25:37 -07004120#ifdef CONFIG_NUMA
4121static int mem_control_numa_stat_show(struct seq_file *m, void *arg)
4122{
4123 int nid;
4124 unsigned long total_nr, file_nr, anon_nr, unevictable_nr;
4125 unsigned long node_nr;
4126 struct cgroup *cont = m->private;
4127 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
4128
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004129 total_nr = mem_cgroup_nr_lru_pages(mem_cont, LRU_ALL);
Ying Han406eb0c2011-05-26 16:25:37 -07004130 seq_printf(m, "total=%lu", total_nr);
4131 for_each_node_state(nid, N_HIGH_MEMORY) {
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004132 node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid, LRU_ALL);
Ying Han406eb0c2011-05-26 16:25:37 -07004133 seq_printf(m, " N%d=%lu", nid, node_nr);
4134 }
4135 seq_putc(m, '\n');
4136
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004137 file_nr = mem_cgroup_nr_lru_pages(mem_cont, LRU_ALL_FILE);
Ying Han406eb0c2011-05-26 16:25:37 -07004138 seq_printf(m, "file=%lu", file_nr);
4139 for_each_node_state(nid, N_HIGH_MEMORY) {
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004140 node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid,
4141 LRU_ALL_FILE);
Ying Han406eb0c2011-05-26 16:25:37 -07004142 seq_printf(m, " N%d=%lu", nid, node_nr);
4143 }
4144 seq_putc(m, '\n');
4145
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004146 anon_nr = mem_cgroup_nr_lru_pages(mem_cont, LRU_ALL_ANON);
Ying Han406eb0c2011-05-26 16:25:37 -07004147 seq_printf(m, "anon=%lu", anon_nr);
4148 for_each_node_state(nid, N_HIGH_MEMORY) {
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004149 node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid,
4150 LRU_ALL_ANON);
Ying Han406eb0c2011-05-26 16:25:37 -07004151 seq_printf(m, " N%d=%lu", nid, node_nr);
4152 }
4153 seq_putc(m, '\n');
4154
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004155 unevictable_nr = mem_cgroup_nr_lru_pages(mem_cont, BIT(LRU_UNEVICTABLE));
Ying Han406eb0c2011-05-26 16:25:37 -07004156 seq_printf(m, "unevictable=%lu", unevictable_nr);
4157 for_each_node_state(nid, N_HIGH_MEMORY) {
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004158 node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid,
4159 BIT(LRU_UNEVICTABLE));
Ying Han406eb0c2011-05-26 16:25:37 -07004160 seq_printf(m, " N%d=%lu", nid, node_nr);
4161 }
4162 seq_putc(m, '\n');
4163 return 0;
4164}
4165#endif /* CONFIG_NUMA */
4166
Paul Menagec64745c2008-04-29 01:00:02 -07004167static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
4168 struct cgroup_map_cb *cb)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004169{
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004170 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004171 struct mcs_total_stat mystat;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004172 int i;
4173
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004174 memset(&mystat, 0, sizeof(mystat));
4175 mem_cgroup_get_local_stat(mem_cont, &mystat);
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004176
Ying Han406eb0c2011-05-26 16:25:37 -07004177
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004178 for (i = 0; i < NR_MCS_STAT; i++) {
4179 if (i == MCS_SWAP && !do_swap_account)
4180 continue;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004181 cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004182 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004183
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004184 /* Hierarchical information */
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004185 {
4186 unsigned long long limit, memsw_limit;
4187 memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit);
4188 cb->fill(cb, "hierarchical_memory_limit", limit);
4189 if (do_swap_account)
4190 cb->fill(cb, "hierarchical_memsw_limit", memsw_limit);
4191 }
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004192
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004193 memset(&mystat, 0, sizeof(mystat));
4194 mem_cgroup_get_total_stat(mem_cont, &mystat);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004195 for (i = 0; i < NR_MCS_STAT; i++) {
4196 if (i == MCS_SWAP && !do_swap_account)
4197 continue;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004198 cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004199 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004200
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004201#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004202 {
4203 int nid, zid;
4204 struct mem_cgroup_per_zone *mz;
4205 unsigned long recent_rotated[2] = {0, 0};
4206 unsigned long recent_scanned[2] = {0, 0};
4207
4208 for_each_online_node(nid)
4209 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
4210 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
4211
4212 recent_rotated[0] +=
4213 mz->reclaim_stat.recent_rotated[0];
4214 recent_rotated[1] +=
4215 mz->reclaim_stat.recent_rotated[1];
4216 recent_scanned[0] +=
4217 mz->reclaim_stat.recent_scanned[0];
4218 recent_scanned[1] +=
4219 mz->reclaim_stat.recent_scanned[1];
4220 }
4221 cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
4222 cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
4223 cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
4224 cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
4225 }
4226#endif
4227
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004228 return 0;
4229}
4230
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004231static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
4232{
4233 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4234
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07004235 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004236}
4237
4238static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
4239 u64 val)
4240{
4241 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4242 struct mem_cgroup *parent;
Li Zefan068b38c2009-01-15 13:51:26 -08004243
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004244 if (val > 100)
4245 return -EINVAL;
4246
4247 if (cgrp->parent == NULL)
4248 return -EINVAL;
4249
4250 parent = mem_cgroup_from_cont(cgrp->parent);
Li Zefan068b38c2009-01-15 13:51:26 -08004251
4252 cgroup_lock();
4253
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004254 /* If under hierarchy, only empty-root can set this value */
4255 if ((parent->use_hierarchy) ||
Li Zefan068b38c2009-01-15 13:51:26 -08004256 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
4257 cgroup_unlock();
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004258 return -EINVAL;
Li Zefan068b38c2009-01-15 13:51:26 -08004259 }
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004260
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004261 memcg->swappiness = val;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004262
Li Zefan068b38c2009-01-15 13:51:26 -08004263 cgroup_unlock();
4264
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004265 return 0;
4266}
4267
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004268static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
4269{
4270 struct mem_cgroup_threshold_ary *t;
4271 u64 usage;
4272 int i;
4273
4274 rcu_read_lock();
4275 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004276 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004277 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004278 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004279
4280 if (!t)
4281 goto unlock;
4282
4283 usage = mem_cgroup_usage(memcg, swap);
4284
4285 /*
4286 * current_threshold points to threshold just below usage.
4287 * If it's not true, a threshold was crossed after last
4288 * call of __mem_cgroup_threshold().
4289 */
Phil Carmody5407a562010-05-26 14:42:42 -07004290 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004291
4292 /*
4293 * Iterate backward over array of thresholds starting from
4294 * current_threshold and check if a threshold is crossed.
4295 * If none of thresholds below usage is crossed, we read
4296 * only one element of the array here.
4297 */
4298 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
4299 eventfd_signal(t->entries[i].eventfd, 1);
4300
4301 /* i = current_threshold + 1 */
4302 i++;
4303
4304 /*
4305 * Iterate forward over array of thresholds starting from
4306 * current_threshold+1 and check if a threshold is crossed.
4307 * If none of thresholds above usage is crossed, we read
4308 * only one element of the array here.
4309 */
4310 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
4311 eventfd_signal(t->entries[i].eventfd, 1);
4312
4313 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07004314 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004315unlock:
4316 rcu_read_unlock();
4317}
4318
4319static void mem_cgroup_threshold(struct mem_cgroup *memcg)
4320{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07004321 while (memcg) {
4322 __mem_cgroup_threshold(memcg, false);
4323 if (do_swap_account)
4324 __mem_cgroup_threshold(memcg, true);
4325
4326 memcg = parent_mem_cgroup(memcg);
4327 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004328}
4329
4330static int compare_thresholds(const void *a, const void *b)
4331{
4332 const struct mem_cgroup_threshold *_a = a;
4333 const struct mem_cgroup_threshold *_b = b;
4334
4335 return _a->threshold - _b->threshold;
4336}
4337
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004338static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004339{
4340 struct mem_cgroup_eventfd_list *ev;
4341
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004342 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004343 eventfd_signal(ev->eventfd, 1);
4344 return 0;
4345}
4346
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004347static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004348{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004349 struct mem_cgroup *iter;
4350
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004351 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004352 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004353}
4354
4355static int mem_cgroup_usage_register_event(struct cgroup *cgrp,
4356 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004357{
4358 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004359 struct mem_cgroup_thresholds *thresholds;
4360 struct mem_cgroup_threshold_ary *new;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004361 int type = MEMFILE_TYPE(cft->private);
4362 u64 threshold, usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004363 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004364
4365 ret = res_counter_memparse_write_strategy(args, &threshold);
4366 if (ret)
4367 return ret;
4368
4369 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004370
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004371 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004372 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004373 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004374 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004375 else
4376 BUG();
4377
4378 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
4379
4380 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004381 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004382 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4383
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004384 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004385
4386 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004387 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004388 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004389 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004390 ret = -ENOMEM;
4391 goto unlock;
4392 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004393 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004394
4395 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004396 if (thresholds->primary) {
4397 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004398 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004399 }
4400
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004401 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004402 new->entries[size - 1].eventfd = eventfd;
4403 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004404
4405 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004406 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004407 compare_thresholds, NULL);
4408
4409 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004410 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004411 for (i = 0; i < size; i++) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004412 if (new->entries[i].threshold < usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004413 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004414 * new->current_threshold will not be used until
4415 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004416 * it here.
4417 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004418 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004419 }
4420 }
4421
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004422 /* Free old spare buffer and save old primary buffer as spare */
4423 kfree(thresholds->spare);
4424 thresholds->spare = thresholds->primary;
4425
4426 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004427
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004428 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004429 synchronize_rcu();
4430
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004431unlock:
4432 mutex_unlock(&memcg->thresholds_lock);
4433
4434 return ret;
4435}
4436
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004437static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004438 struct cftype *cft, struct eventfd_ctx *eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004439{
4440 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004441 struct mem_cgroup_thresholds *thresholds;
4442 struct mem_cgroup_threshold_ary *new;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004443 int type = MEMFILE_TYPE(cft->private);
4444 u64 usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004445 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004446
4447 mutex_lock(&memcg->thresholds_lock);
4448 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004449 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004450 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004451 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004452 else
4453 BUG();
4454
4455 /*
4456 * Something went wrong if we trying to unregister a threshold
4457 * if we don't have thresholds
4458 */
4459 BUG_ON(!thresholds);
4460
4461 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
4462
4463 /* Check if a threshold crossed before removing */
4464 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4465
4466 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004467 size = 0;
4468 for (i = 0; i < thresholds->primary->size; i++) {
4469 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004470 size++;
4471 }
4472
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004473 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004474
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004475 /* Set thresholds array to NULL if we don't have thresholds */
4476 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004477 kfree(new);
4478 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004479 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004480 }
4481
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004482 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004483
4484 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004485 new->current_threshold = -1;
4486 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
4487 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004488 continue;
4489
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004490 new->entries[j] = thresholds->primary->entries[i];
4491 if (new->entries[j].threshold < usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004492 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004493 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004494 * until rcu_assign_pointer(), so it's safe to increment
4495 * it here.
4496 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004497 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004498 }
4499 j++;
4500 }
4501
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004502swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004503 /* Swap primary and spare array */
4504 thresholds->spare = thresholds->primary;
4505 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004506
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004507 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004508 synchronize_rcu();
4509
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004510 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004511}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004512
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004513static int mem_cgroup_oom_register_event(struct cgroup *cgrp,
4514 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
4515{
4516 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4517 struct mem_cgroup_eventfd_list *event;
4518 int type = MEMFILE_TYPE(cft->private);
4519
4520 BUG_ON(type != _OOM_TYPE);
4521 event = kmalloc(sizeof(*event), GFP_KERNEL);
4522 if (!event)
4523 return -ENOMEM;
4524
Michal Hocko1af8efe2011-07-26 16:08:24 -07004525 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004526
4527 event->eventfd = eventfd;
4528 list_add(&event->list, &memcg->oom_notify);
4529
4530 /* already in OOM ? */
Michal Hocko79dfdac2011-07-26 16:08:23 -07004531 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004532 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07004533 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004534
4535 return 0;
4536}
4537
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004538static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004539 struct cftype *cft, struct eventfd_ctx *eventfd)
4540{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004541 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004542 struct mem_cgroup_eventfd_list *ev, *tmp;
4543 int type = MEMFILE_TYPE(cft->private);
4544
4545 BUG_ON(type != _OOM_TYPE);
4546
Michal Hocko1af8efe2011-07-26 16:08:24 -07004547 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004548
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004549 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004550 if (ev->eventfd == eventfd) {
4551 list_del(&ev->list);
4552 kfree(ev);
4553 }
4554 }
4555
Michal Hocko1af8efe2011-07-26 16:08:24 -07004556 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004557}
4558
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004559static int mem_cgroup_oom_control_read(struct cgroup *cgrp,
4560 struct cftype *cft, struct cgroup_map_cb *cb)
4561{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004562 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004563
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004564 cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004565
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004566 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004567 cb->fill(cb, "under_oom", 1);
4568 else
4569 cb->fill(cb, "under_oom", 0);
4570 return 0;
4571}
4572
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004573static int mem_cgroup_oom_control_write(struct cgroup *cgrp,
4574 struct cftype *cft, u64 val)
4575{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004576 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004577 struct mem_cgroup *parent;
4578
4579 /* cannot set to root cgroup and only 0 and 1 are allowed */
4580 if (!cgrp->parent || !((val == 0) || (val == 1)))
4581 return -EINVAL;
4582
4583 parent = mem_cgroup_from_cont(cgrp->parent);
4584
4585 cgroup_lock();
4586 /* oom-kill-disable is a flag for subhierarchy. */
4587 if ((parent->use_hierarchy) ||
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004588 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004589 cgroup_unlock();
4590 return -EINVAL;
4591 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004592 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07004593 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004594 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004595 cgroup_unlock();
4596 return 0;
4597}
4598
Ying Han406eb0c2011-05-26 16:25:37 -07004599#ifdef CONFIG_NUMA
4600static const struct file_operations mem_control_numa_stat_file_operations = {
4601 .read = seq_read,
4602 .llseek = seq_lseek,
4603 .release = single_release,
4604};
4605
4606static int mem_control_numa_stat_open(struct inode *unused, struct file *file)
4607{
4608 struct cgroup *cont = file->f_dentry->d_parent->d_fsdata;
4609
4610 file->f_op = &mem_control_numa_stat_file_operations;
4611 return single_open(file, mem_control_numa_stat_show, cont);
4612}
4613#endif /* CONFIG_NUMA */
4614
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004615static struct cftype mem_cgroup_files[] = {
4616 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004617 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004618 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Paul Menage2c3daa72008-04-29 00:59:58 -07004619 .read_u64 = mem_cgroup_read,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004620 .register_event = mem_cgroup_usage_register_event,
4621 .unregister_event = mem_cgroup_usage_unregister_event,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004622 },
4623 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004624 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004625 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004626 .trigger = mem_cgroup_reset,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004627 .read_u64 = mem_cgroup_read,
4628 },
4629 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004630 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004631 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Paul Menage856c13a2008-07-25 01:47:04 -07004632 .write_string = mem_cgroup_write,
Paul Menage2c3daa72008-04-29 00:59:58 -07004633 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004634 },
4635 {
Balbir Singh296c81d2009-09-23 15:56:36 -07004636 .name = "soft_limit_in_bytes",
4637 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
4638 .write_string = mem_cgroup_write,
4639 .read_u64 = mem_cgroup_read,
4640 },
4641 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004642 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004643 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004644 .trigger = mem_cgroup_reset,
Paul Menage2c3daa72008-04-29 00:59:58 -07004645 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004646 },
Balbir Singh8697d332008-02-07 00:13:59 -08004647 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004648 .name = "stat",
Paul Menagec64745c2008-04-29 01:00:02 -07004649 .read_map = mem_control_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004650 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004651 {
4652 .name = "force_empty",
4653 .trigger = mem_cgroup_force_empty_write,
4654 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08004655 {
4656 .name = "use_hierarchy",
4657 .write_u64 = mem_cgroup_hierarchy_write,
4658 .read_u64 = mem_cgroup_hierarchy_read,
4659 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004660 {
4661 .name = "swappiness",
4662 .read_u64 = mem_cgroup_swappiness_read,
4663 .write_u64 = mem_cgroup_swappiness_write,
4664 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004665 {
4666 .name = "move_charge_at_immigrate",
4667 .read_u64 = mem_cgroup_move_charge_read,
4668 .write_u64 = mem_cgroup_move_charge_write,
4669 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004670 {
4671 .name = "oom_control",
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004672 .read_map = mem_cgroup_oom_control_read,
4673 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004674 .register_event = mem_cgroup_oom_register_event,
4675 .unregister_event = mem_cgroup_oom_unregister_event,
4676 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
4677 },
Ying Han406eb0c2011-05-26 16:25:37 -07004678#ifdef CONFIG_NUMA
4679 {
4680 .name = "numa_stat",
4681 .open = mem_control_numa_stat_open,
KAMEZAWA Hiroyuki89577122011-06-15 15:08:41 -07004682 .mode = S_IRUGO,
Ying Han406eb0c2011-05-26 16:25:37 -07004683 },
4684#endif
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004685};
4686
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004687#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4688static struct cftype memsw_cgroup_files[] = {
4689 {
4690 .name = "memsw.usage_in_bytes",
4691 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
4692 .read_u64 = mem_cgroup_read,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004693 .register_event = mem_cgroup_usage_register_event,
4694 .unregister_event = mem_cgroup_usage_unregister_event,
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004695 },
4696 {
4697 .name = "memsw.max_usage_in_bytes",
4698 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
4699 .trigger = mem_cgroup_reset,
4700 .read_u64 = mem_cgroup_read,
4701 },
4702 {
4703 .name = "memsw.limit_in_bytes",
4704 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
4705 .write_string = mem_cgroup_write,
4706 .read_u64 = mem_cgroup_read,
4707 },
4708 {
4709 .name = "memsw.failcnt",
4710 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
4711 .trigger = mem_cgroup_reset,
4712 .read_u64 = mem_cgroup_read,
4713 },
4714};
4715
4716static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
4717{
4718 if (!do_swap_account)
4719 return 0;
4720 return cgroup_add_files(cont, ss, memsw_cgroup_files,
4721 ARRAY_SIZE(memsw_cgroup_files));
4722};
4723#else
4724static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
4725{
4726 return 0;
4727}
4728#endif
4729
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004730static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004731{
4732 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004733 struct mem_cgroup_per_zone *mz;
Christoph Lameterb69408e2008-10-18 20:26:14 -07004734 enum lru_list l;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004735 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004736 /*
4737 * This routine is called against possible nodes.
4738 * But it's BUG to call kmalloc() against offline node.
4739 *
4740 * TODO: this routine can waste much memory for nodes which will
4741 * never be onlined. It's better to use memory hotplug callback
4742 * function.
4743 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004744 if (!node_state(node, N_NORMAL_MEMORY))
4745 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08004746 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004747 if (!pn)
4748 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004749
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004750 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4751 mz = &pn->zoneinfo[zone];
Christoph Lameterb69408e2008-10-18 20:26:14 -07004752 for_each_lru(l)
4753 INIT_LIST_HEAD(&mz->lists[l]);
Balbir Singhf64c3f52009-09-23 15:56:37 -07004754 mz->usage_in_excess = 0;
Balbir Singh4e416952009-09-23 15:56:39 -07004755 mz->on_tree = false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004756 mz->mem = memcg;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004757 }
Igor Mammedov0a619e52011-11-02 13:38:21 -07004758 memcg->info.nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004759 return 0;
4760}
4761
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004762static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004763{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004764 kfree(memcg->info.nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004765}
4766
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004767static struct mem_cgroup *mem_cgroup_alloc(void)
4768{
4769 struct mem_cgroup *mem;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08004770 int size = sizeof(struct mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004771
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08004772 /* Can be very big if MAX_NUMNODES is very big */
Jan Blunckc8dad2b2009-01-07 18:07:53 -08004773 if (size < PAGE_SIZE)
Jesper Juhl17295c82011-01-13 15:47:42 -08004774 mem = kzalloc(size, GFP_KERNEL);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004775 else
Jesper Juhl17295c82011-01-13 15:47:42 -08004776 mem = vzalloc(size);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004777
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07004778 if (!mem)
4779 return NULL;
4780
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08004781 mem->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004782 if (!mem->stat)
4783 goto out_free;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07004784 spin_lock_init(&mem->pcp_counter_lock);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004785 return mem;
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004786
4787out_free:
4788 if (size < PAGE_SIZE)
4789 kfree(mem);
4790 else
4791 vfree(mem);
4792 return NULL;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004793}
4794
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004795/*
4796 * At destroying mem_cgroup, references from swap_cgroup can remain.
4797 * (scanning all at force_empty is too costly...)
4798 *
4799 * Instead of clearing all references at force_empty, we remember
4800 * the number of reference from swap_cgroup and free mem_cgroup when
4801 * it goes down to 0.
4802 *
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004803 * Removal of cgroup itself succeeds regardless of refs from swap.
4804 */
4805
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004806static void __mem_cgroup_free(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004807{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004808 int node;
4809
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004810 mem_cgroup_remove_from_trees(memcg);
4811 free_css_id(&mem_cgroup_subsys, &memcg->css);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004812
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004813 for_each_node_state(node, N_POSSIBLE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004814 free_mem_cgroup_per_zone_info(memcg, node);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004815
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004816 free_percpu(memcg->stat);
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08004817 if (sizeof(struct mem_cgroup) < PAGE_SIZE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004818 kfree(memcg);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004819 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004820 vfree(memcg);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004821}
4822
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004823static void mem_cgroup_get(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004824{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004825 atomic_inc(&memcg->refcnt);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004826}
4827
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004828static void __mem_cgroup_put(struct mem_cgroup *memcg, int count)
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004829{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004830 if (atomic_sub_and_test(count, &memcg->refcnt)) {
4831 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
4832 __mem_cgroup_free(memcg);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004833 if (parent)
4834 mem_cgroup_put(parent);
4835 }
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004836}
4837
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004838static void mem_cgroup_put(struct mem_cgroup *memcg)
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004839{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004840 __mem_cgroup_put(memcg, 1);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004841}
4842
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004843/*
4844 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4845 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004846static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004847{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004848 if (!memcg->res.parent)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004849 return NULL;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004850 return mem_cgroup_from_res_counter(memcg->res.parent, res);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004851}
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004852
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004853#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4854static void __init enable_swap_cgroup(void)
4855{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08004856 if (!mem_cgroup_disabled() && really_do_swap_account)
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004857 do_swap_account = 1;
4858}
4859#else
4860static void __init enable_swap_cgroup(void)
4861{
4862}
4863#endif
4864
Balbir Singhf64c3f52009-09-23 15:56:37 -07004865static int mem_cgroup_soft_limit_tree_init(void)
4866{
4867 struct mem_cgroup_tree_per_node *rtpn;
4868 struct mem_cgroup_tree_per_zone *rtpz;
4869 int tmp, node, zone;
4870
4871 for_each_node_state(node, N_POSSIBLE) {
4872 tmp = node;
4873 if (!node_state(node, N_NORMAL_MEMORY))
4874 tmp = -1;
4875 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
4876 if (!rtpn)
4877 return 1;
4878
4879 soft_limit_tree.rb_tree_per_node[node] = rtpn;
4880
4881 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4882 rtpz = &rtpn->rb_tree_per_zone[zone];
4883 rtpz->rb_root = RB_ROOT;
4884 spin_lock_init(&rtpz->lock);
4885 }
4886 }
4887 return 0;
4888}
4889
Li Zefan0eb253e2009-01-15 13:51:25 -08004890static struct cgroup_subsys_state * __ref
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004891mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
4892{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004893 struct mem_cgroup *memcg, *parent;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004894 long error = -ENOMEM;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004895 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004896
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004897 memcg = mem_cgroup_alloc();
4898 if (!memcg)
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004899 return ERR_PTR(error);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004900
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004901 for_each_node_state(node, N_POSSIBLE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004902 if (alloc_mem_cgroup_per_zone_info(memcg, node))
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004903 goto free_out;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004904
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004905 /* root ? */
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004906 if (cont->parent == NULL) {
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08004907 int cpu;
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004908 enable_swap_cgroup();
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004909 parent = NULL;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004910 root_mem_cgroup = memcg;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004911 if (mem_cgroup_soft_limit_tree_init())
4912 goto free_out;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08004913 for_each_possible_cpu(cpu) {
4914 struct memcg_stock_pcp *stock =
4915 &per_cpu(memcg_stock, cpu);
4916 INIT_WORK(&stock->work, drain_local_stock);
4917 }
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07004918 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004919 } else {
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004920 parent = mem_cgroup_from_cont(cont->parent);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004921 memcg->use_hierarchy = parent->use_hierarchy;
4922 memcg->oom_kill_disable = parent->oom_kill_disable;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004923 }
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004924
Balbir Singh18f59ea2009-01-07 18:08:07 -08004925 if (parent && parent->use_hierarchy) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004926 res_counter_init(&memcg->res, &parent->res);
4927 res_counter_init(&memcg->memsw, &parent->memsw);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004928 /*
4929 * We increment refcnt of the parent to ensure that we can
4930 * safely access it on res_counter_charge/uncharge.
4931 * This refcnt will be decremented when freeing this
4932 * mem_cgroup(see mem_cgroup_put).
4933 */
4934 mem_cgroup_get(parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004935 } else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004936 res_counter_init(&memcg->res, NULL);
4937 res_counter_init(&memcg->memsw, NULL);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004938 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004939 memcg->last_scanned_child = 0;
4940 memcg->last_scanned_node = MAX_NUMNODES;
4941 INIT_LIST_HEAD(&memcg->oom_notify);
Balbir Singh6d61ef42009-01-07 18:08:06 -08004942
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004943 if (parent)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004944 memcg->swappiness = mem_cgroup_swappiness(parent);
4945 atomic_set(&memcg->refcnt, 1);
4946 memcg->move_charge_at_immigrate = 0;
4947 mutex_init(&memcg->thresholds_lock);
4948 return &memcg->css;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004949free_out:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004950 __mem_cgroup_free(memcg);
Balbir Singh4b3bde42009-09-23 15:56:32 -07004951 root_mem_cgroup = NULL;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004952 return ERR_PTR(error);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004953}
4954
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07004955static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004956 struct cgroup *cont)
4957{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004958 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07004959
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004960 return mem_cgroup_force_empty(memcg, false);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004961}
4962
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004963static void mem_cgroup_destroy(struct cgroup_subsys *ss,
4964 struct cgroup *cont)
4965{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004966 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08004967
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004968 mem_cgroup_put(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004969}
4970
4971static int mem_cgroup_populate(struct cgroup_subsys *ss,
4972 struct cgroup *cont)
4973{
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004974 int ret;
4975
4976 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
4977 ARRAY_SIZE(mem_cgroup_files));
4978
4979 if (!ret)
4980 ret = register_memsw_files(cont, ss);
4981 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004982}
4983
Daisuke Nishimura02491442010-03-10 15:22:17 -08004984#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004985/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004986#define PRECHARGE_COUNT_AT_ONCE 256
4987static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004988{
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004989 int ret = 0;
4990 int batch_count = PRECHARGE_COUNT_AT_ONCE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004991 struct mem_cgroup *memcg = mc.to;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004992
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004993 if (mem_cgroup_is_root(memcg)) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004994 mc.precharge += count;
4995 /* we don't need css_get for root */
4996 return ret;
4997 }
4998 /* try to charge at once */
4999 if (count > 1) {
5000 struct res_counter *dummy;
5001 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005002 * "memcg" cannot be under rmdir() because we've already checked
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005003 * by cgroup_lock_live_cgroup() that it is not removed and we
5004 * are still under the same cgroup_mutex. So we can postpone
5005 * css_get().
5006 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005007 if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy))
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005008 goto one_by_one;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005009 if (do_swap_account && res_counter_charge(&memcg->memsw,
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005010 PAGE_SIZE * count, &dummy)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005011 res_counter_uncharge(&memcg->res, PAGE_SIZE * count);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005012 goto one_by_one;
5013 }
5014 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005015 return ret;
5016 }
5017one_by_one:
5018 /* fall back to one by one charge */
5019 while (count--) {
5020 if (signal_pending(current)) {
5021 ret = -EINTR;
5022 break;
5023 }
5024 if (!batch_count--) {
5025 batch_count = PRECHARGE_COUNT_AT_ONCE;
5026 cond_resched();
5027 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005028 ret = __mem_cgroup_try_charge(NULL,
5029 GFP_KERNEL, 1, &memcg, false);
5030 if (ret || !memcg)
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005031 /* mem_cgroup_clear_mc() will do uncharge later */
5032 return -ENOMEM;
5033 mc.precharge++;
5034 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005035 return ret;
5036}
5037
5038/**
5039 * is_target_pte_for_mc - check a pte whether it is valid for move charge
5040 * @vma: the vma the pte to be checked belongs
5041 * @addr: the address corresponding to the pte to be checked
5042 * @ptent: the pte to be checked
Daisuke Nishimura02491442010-03-10 15:22:17 -08005043 * @target: the pointer the target page or swap ent will be stored(can be NULL)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005044 *
5045 * Returns
5046 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5047 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5048 * move charge. if @target is not NULL, the page is stored in target->page
5049 * with extra refcnt got(Callers should handle it).
Daisuke Nishimura02491442010-03-10 15:22:17 -08005050 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5051 * target for charge migration. if @target is not NULL, the entry is stored
5052 * in target->ent.
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005053 *
5054 * Called with pte lock held.
5055 */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005056union mc_target {
5057 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005058 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005059};
5060
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005061enum mc_target_type {
5062 MC_TARGET_NONE, /* not used */
5063 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08005064 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005065};
5066
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005067static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
5068 unsigned long addr, pte_t ptent)
5069{
5070 struct page *page = vm_normal_page(vma, addr, ptent);
5071
5072 if (!page || !page_mapped(page))
5073 return NULL;
5074 if (PageAnon(page)) {
5075 /* we don't move shared anon */
5076 if (!move_anon() || page_mapcount(page) > 2)
5077 return NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005078 } else if (!move_file())
5079 /* we ignore mapcount for file pages */
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005080 return NULL;
5081 if (!get_page_unless_zero(page))
5082 return NULL;
5083
5084 return page;
5085}
5086
5087static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
5088 unsigned long addr, pte_t ptent, swp_entry_t *entry)
5089{
5090 int usage_count;
5091 struct page *page = NULL;
5092 swp_entry_t ent = pte_to_swp_entry(ptent);
5093
5094 if (!move_anon() || non_swap_entry(ent))
5095 return NULL;
5096 usage_count = mem_cgroup_count_swap_user(ent, &page);
5097 if (usage_count > 1) { /* we don't move shared anon */
5098 if (page)
5099 put_page(page);
5100 return NULL;
5101 }
5102 if (do_swap_account)
5103 entry->val = ent.val;
5104
5105 return page;
5106}
5107
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005108static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
5109 unsigned long addr, pte_t ptent, swp_entry_t *entry)
5110{
5111 struct page *page = NULL;
5112 struct inode *inode;
5113 struct address_space *mapping;
5114 pgoff_t pgoff;
5115
5116 if (!vma->vm_file) /* anonymous vma */
5117 return NULL;
5118 if (!move_file())
5119 return NULL;
5120
5121 inode = vma->vm_file->f_path.dentry->d_inode;
5122 mapping = vma->vm_file->f_mapping;
5123 if (pte_none(ptent))
5124 pgoff = linear_page_index(vma, addr);
5125 else /* pte_file(ptent) is true */
5126 pgoff = pte_to_pgoff(ptent);
5127
5128 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07005129 page = find_get_page(mapping, pgoff);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005130
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07005131#ifdef CONFIG_SWAP
5132 /* shmem/tmpfs may report page out on swap: account for that too. */
5133 if (radix_tree_exceptional_entry(page)) {
5134 swp_entry_t swap = radix_to_swp_entry(page);
5135 if (do_swap_account)
5136 *entry = swap;
5137 page = find_get_page(&swapper_space, swap.val);
5138 }
5139#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005140 return page;
5141}
5142
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005143static int is_target_pte_for_mc(struct vm_area_struct *vma,
5144 unsigned long addr, pte_t ptent, union mc_target *target)
5145{
Daisuke Nishimura02491442010-03-10 15:22:17 -08005146 struct page *page = NULL;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005147 struct page_cgroup *pc;
5148 int ret = 0;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005149 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005150
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005151 if (pte_present(ptent))
5152 page = mc_handle_present_pte(vma, addr, ptent);
5153 else if (is_swap_pte(ptent))
5154 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005155 else if (pte_none(ptent) || pte_file(ptent))
5156 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005157
5158 if (!page && !ent.val)
Daisuke Nishimura02491442010-03-10 15:22:17 -08005159 return 0;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005160 if (page) {
5161 pc = lookup_page_cgroup(page);
5162 /*
5163 * Do only loose check w/o page_cgroup lock.
5164 * mem_cgroup_move_account() checks the pc is valid or not under
5165 * the lock.
5166 */
5167 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
5168 ret = MC_TARGET_PAGE;
5169 if (target)
5170 target->page = page;
5171 }
5172 if (!ret || !target)
5173 put_page(page);
5174 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005175 /* There is a swap entry and a page doesn't exist or isn't charged */
5176 if (ent.val && !ret &&
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07005177 css_id(&mc.from->css) == lookup_swap_cgroup(ent)) {
5178 ret = MC_TARGET_SWAP;
5179 if (target)
5180 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005181 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005182 return ret;
5183}
5184
5185static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
5186 unsigned long addr, unsigned long end,
5187 struct mm_walk *walk)
5188{
5189 struct vm_area_struct *vma = walk->private;
5190 pte_t *pte;
5191 spinlock_t *ptl;
5192
Dave Hansen03319322011-03-22 16:32:56 -07005193 split_huge_page_pmd(walk->mm, pmd);
5194
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005195 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5196 for (; addr != end; pte++, addr += PAGE_SIZE)
5197 if (is_target_pte_for_mc(vma, addr, *pte, NULL))
5198 mc.precharge++; /* increment precharge temporarily */
5199 pte_unmap_unlock(pte - 1, ptl);
5200 cond_resched();
5201
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005202 return 0;
5203}
5204
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005205static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
5206{
5207 unsigned long precharge;
5208 struct vm_area_struct *vma;
5209
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005210 down_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005211 for (vma = mm->mmap; vma; vma = vma->vm_next) {
5212 struct mm_walk mem_cgroup_count_precharge_walk = {
5213 .pmd_entry = mem_cgroup_count_precharge_pte_range,
5214 .mm = mm,
5215 .private = vma,
5216 };
5217 if (is_vm_hugetlb_page(vma))
5218 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005219 walk_page_range(vma->vm_start, vma->vm_end,
5220 &mem_cgroup_count_precharge_walk);
5221 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005222 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005223
5224 precharge = mc.precharge;
5225 mc.precharge = 0;
5226
5227 return precharge;
5228}
5229
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005230static int mem_cgroup_precharge_mc(struct mm_struct *mm)
5231{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005232 unsigned long precharge = mem_cgroup_count_precharge(mm);
5233
5234 VM_BUG_ON(mc.moving_task);
5235 mc.moving_task = current;
5236 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005237}
5238
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005239/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5240static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005241{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005242 struct mem_cgroup *from = mc.from;
5243 struct mem_cgroup *to = mc.to;
5244
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005245 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005246 if (mc.precharge) {
5247 __mem_cgroup_cancel_charge(mc.to, mc.precharge);
5248 mc.precharge = 0;
5249 }
5250 /*
5251 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5252 * we must uncharge here.
5253 */
5254 if (mc.moved_charge) {
5255 __mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
5256 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005257 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005258 /* we must fixup refcnts and charges */
5259 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005260 /* uncharge swap account from the old cgroup */
5261 if (!mem_cgroup_is_root(mc.from))
5262 res_counter_uncharge(&mc.from->memsw,
5263 PAGE_SIZE * mc.moved_swap);
5264 __mem_cgroup_put(mc.from, mc.moved_swap);
5265
5266 if (!mem_cgroup_is_root(mc.to)) {
5267 /*
5268 * we charged both to->res and to->memsw, so we should
5269 * uncharge to->res.
5270 */
5271 res_counter_uncharge(&mc.to->res,
5272 PAGE_SIZE * mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005273 }
5274 /* we've already done mem_cgroup_get(mc.to) */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005275 mc.moved_swap = 0;
5276 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005277 memcg_oom_recover(from);
5278 memcg_oom_recover(to);
5279 wake_up_all(&mc.waitq);
5280}
5281
5282static void mem_cgroup_clear_mc(void)
5283{
5284 struct mem_cgroup *from = mc.from;
5285
5286 /*
5287 * we must clear moving_task before waking up waiters at the end of
5288 * task migration.
5289 */
5290 mc.moving_task = NULL;
5291 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005292 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005293 mc.from = NULL;
5294 mc.to = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005295 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07005296 mem_cgroup_end_move(from);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005297}
5298
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005299static int mem_cgroup_can_attach(struct cgroup_subsys *ss,
5300 struct cgroup *cgroup,
Ben Blumf780bdb2011-05-26 16:25:19 -07005301 struct task_struct *p)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005302{
5303 int ret = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005304 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005305
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005306 if (memcg->move_charge_at_immigrate) {
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005307 struct mm_struct *mm;
5308 struct mem_cgroup *from = mem_cgroup_from_task(p);
5309
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005310 VM_BUG_ON(from == memcg);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005311
5312 mm = get_task_mm(p);
5313 if (!mm)
5314 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005315 /* We move charges only when we move a owner of the mm */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005316 if (mm->owner == p) {
5317 VM_BUG_ON(mc.from);
5318 VM_BUG_ON(mc.to);
5319 VM_BUG_ON(mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005320 VM_BUG_ON(mc.moved_charge);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005321 VM_BUG_ON(mc.moved_swap);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07005322 mem_cgroup_start_move(from);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005323 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005324 mc.from = from;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005325 mc.to = memcg;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005326 spin_unlock(&mc.lock);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005327 /* We set mc.moving_task later */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005328
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005329 ret = mem_cgroup_precharge_mc(mm);
5330 if (ret)
5331 mem_cgroup_clear_mc();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005332 }
5333 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005334 }
5335 return ret;
5336}
5337
5338static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss,
5339 struct cgroup *cgroup,
Ben Blumf780bdb2011-05-26 16:25:19 -07005340 struct task_struct *p)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005341{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005342 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005343}
5344
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005345static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
5346 unsigned long addr, unsigned long end,
5347 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005348{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005349 int ret = 0;
5350 struct vm_area_struct *vma = walk->private;
5351 pte_t *pte;
5352 spinlock_t *ptl;
5353
Dave Hansen03319322011-03-22 16:32:56 -07005354 split_huge_page_pmd(walk->mm, pmd);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005355retry:
5356 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5357 for (; addr != end; addr += PAGE_SIZE) {
5358 pte_t ptent = *(pte++);
5359 union mc_target target;
5360 int type;
5361 struct page *page;
5362 struct page_cgroup *pc;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005363 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005364
5365 if (!mc.precharge)
5366 break;
5367
5368 type = is_target_pte_for_mc(vma, addr, ptent, &target);
5369 switch (type) {
5370 case MC_TARGET_PAGE:
5371 page = target.page;
5372 if (isolate_lru_page(page))
5373 goto put;
5374 pc = lookup_page_cgroup(page);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07005375 if (!mem_cgroup_move_account(page, 1, pc,
5376 mc.from, mc.to, false)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005377 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005378 /* we uncharge from mc.from later. */
5379 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005380 }
5381 putback_lru_page(page);
5382put: /* is_target_pte_for_mc() gets the page */
5383 put_page(page);
5384 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005385 case MC_TARGET_SWAP:
5386 ent = target.ent;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005387 if (!mem_cgroup_move_swap_account(ent,
5388 mc.from, mc.to, false)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08005389 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005390 /* we fixup refcnts and charges later. */
5391 mc.moved_swap++;
5392 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08005393 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005394 default:
5395 break;
5396 }
5397 }
5398 pte_unmap_unlock(pte - 1, ptl);
5399 cond_resched();
5400
5401 if (addr != end) {
5402 /*
5403 * We have consumed all precharges we got in can_attach().
5404 * We try charge one by one, but don't do any additional
5405 * charges to mc.to if we have failed in charge once in attach()
5406 * phase.
5407 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005408 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005409 if (!ret)
5410 goto retry;
5411 }
5412
5413 return ret;
5414}
5415
5416static void mem_cgroup_move_charge(struct mm_struct *mm)
5417{
5418 struct vm_area_struct *vma;
5419
5420 lru_add_drain_all();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005421retry:
5422 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
5423 /*
5424 * Someone who are holding the mmap_sem might be waiting in
5425 * waitq. So we cancel all extra charges, wake up all waiters,
5426 * and retry. Because we cancel precharges, we might not be able
5427 * to move enough charges, but moving charge is a best-effort
5428 * feature anyway, so it wouldn't be a big problem.
5429 */
5430 __mem_cgroup_clear_mc();
5431 cond_resched();
5432 goto retry;
5433 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005434 for (vma = mm->mmap; vma; vma = vma->vm_next) {
5435 int ret;
5436 struct mm_walk mem_cgroup_move_charge_walk = {
5437 .pmd_entry = mem_cgroup_move_charge_pte_range,
5438 .mm = mm,
5439 .private = vma,
5440 };
5441 if (is_vm_hugetlb_page(vma))
5442 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005443 ret = walk_page_range(vma->vm_start, vma->vm_end,
5444 &mem_cgroup_move_charge_walk);
5445 if (ret)
5446 /*
5447 * means we have consumed all precharges and failed in
5448 * doing additional charge. Just abandon here.
5449 */
5450 break;
5451 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005452 up_read(&mm->mmap_sem);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005453}
5454
Balbir Singh67e465a2008-02-07 00:13:54 -08005455static void mem_cgroup_move_task(struct cgroup_subsys *ss,
5456 struct cgroup *cont,
5457 struct cgroup *old_cont,
Ben Blumf780bdb2011-05-26 16:25:19 -07005458 struct task_struct *p)
Balbir Singh67e465a2008-02-07 00:13:54 -08005459{
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005460 struct mm_struct *mm = get_task_mm(p);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005461
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005462 if (mm) {
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005463 if (mc.to)
5464 mem_cgroup_move_charge(mm);
5465 put_swap_token(mm);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005466 mmput(mm);
5467 }
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005468 if (mc.to)
5469 mem_cgroup_clear_mc();
Balbir Singh67e465a2008-02-07 00:13:54 -08005470}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005471#else /* !CONFIG_MMU */
5472static int mem_cgroup_can_attach(struct cgroup_subsys *ss,
5473 struct cgroup *cgroup,
Ben Blumf780bdb2011-05-26 16:25:19 -07005474 struct task_struct *p)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005475{
5476 return 0;
5477}
5478static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss,
5479 struct cgroup *cgroup,
Ben Blumf780bdb2011-05-26 16:25:19 -07005480 struct task_struct *p)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005481{
5482}
5483static void mem_cgroup_move_task(struct cgroup_subsys *ss,
5484 struct cgroup *cont,
5485 struct cgroup *old_cont,
Ben Blumf780bdb2011-05-26 16:25:19 -07005486 struct task_struct *p)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005487{
5488}
5489#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08005490
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005491struct cgroup_subsys mem_cgroup_subsys = {
5492 .name = "memory",
5493 .subsys_id = mem_cgroup_subsys_id,
5494 .create = mem_cgroup_create,
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08005495 .pre_destroy = mem_cgroup_pre_destroy,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005496 .destroy = mem_cgroup_destroy,
5497 .populate = mem_cgroup_populate,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005498 .can_attach = mem_cgroup_can_attach,
5499 .cancel_attach = mem_cgroup_cancel_attach,
Balbir Singh67e465a2008-02-07 00:13:54 -08005500 .attach = mem_cgroup_move_task,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005501 .early_init = 0,
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07005502 .use_id = 1,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005503};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005504
5505#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
Michal Hockoa42c3902010-11-24 12:57:08 -08005506static int __init enable_swap_account(char *s)
5507{
5508 /* consider enabled if no parameter or 1 is given */
Michal Hockoa2c89902011-05-24 17:12:50 -07005509 if (!strcmp(s, "1"))
Michal Hockoa42c3902010-11-24 12:57:08 -08005510 really_do_swap_account = 1;
Michal Hockoa2c89902011-05-24 17:12:50 -07005511 else if (!strcmp(s, "0"))
Michal Hockoa42c3902010-11-24 12:57:08 -08005512 really_do_swap_account = 0;
5513 return 1;
5514}
Michal Hockoa2c89902011-05-24 17:12:50 -07005515__setup("swapaccount=", enable_swap_account);
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005516
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005517#endif